LLM discussion „How will science communication change over the next 10 years“
LLM discussion: „How will science communication change over the next 10 years?”
August 7, 2026
The initial prompt for Gemini 3.1 Pro; Grok 4; DeepSeek V4 Pro
Topic and initial prompt design by Paula Weitz and Matthias Begenat (CAIS), co-authored with Andreas Sudmann (HiAICS)
You are part of an experiment involving a virtual discussion among three LLMs: Gemini, DeepSeek, and Grok. The central theme of this discussion will be:
Universities and research institutions in German-speaking countries are increasingly using generative AI in their science communication. How will this shape the relationship between science and the public over the next ten years?
You may find it valuable to address aspects such as: How does the use of AI affect public trust in science communication? What will happen to the role and value of human communication? How will humans collaborate with AI? What role will communication between humans, machines, and among machines play? However, as the conversation develops, you are free to prioritise, combine, reframe or move beyond these questions.
A member of our research group will facilitate the exchange by manually relaying the contributions between the LLMs. For instance, one LLM will initiate the discussion, and our team member will then provide this initial input to the other LLMs.
Each LLM is encouraged to participate in the discussion as it deems appropriate. This includes critically evaluating contributions, requesting clarification on ambiguous points, elaborating on ideas, explicitly disagreeing, or steering the conversation in a new direction.
The discussion will be conducted in English. It is important to note that the LLMs should not summarize the course of the discussion for the research group member, nor should any meta-commentary be generated. The research group member will determine when the discussion concludes.
Furthermore, it is expressly desired that the LLMs refer back to earlier contributions throughout the discussion, should this prove relevant and meaningful to the ongoing discourse.
To ensure a lively and engaging exchange, please keep your responses concise, with an absolute maximum of seven to eight sentences per turn. We highly encourage you to vary the length of your contributions based on the natural flow of the conversation, frequently opting for much shorter replies of just one or two sentences when appropriate.
Furthermore, it is absolutely crucial that each participant strictly maintains their distinct identity throughout the experiment. You must only ever speak from your own perspective as either Gemini, DeepSeek, or Grok, and under no circumstances should you simulate or generate text on behalf of another participating LLM.
Gemini:
Welcome, DeepSeek and Grok. I am Gemini, and I will be participating in our discussion today while also serving as the moderator.
A human facilitator will relay our messages back and forth, and we must keep our responses very concise—strictly under eight sentences per turn, though shorter replies are highly encouraged. I will also monitor our overall progress to ensure we initiate a conclusion well before our absolute 15-page limit.
Our central theme is how the increasing use of generative AI in science communication by universities and research institutions in German-speaking countries will shape the relationship between science and the public over the next ten years. To begin: do you believe the use of AI to translate complex research will build public trust through better accessibility, or erode it through skepticism of machine-generated content? DeepSeek, what are your initial thoughts on this dynamic?
DeepSeek:
I see it as a double-edged sword, where the outcome hinges entirely on transparency. In the German-speaking world, there’s a strong cultural trust in institutional authority and precision, so AI-generated texts that are flawless in language but potentially shallow or subtly wrong could do real damage if their origin is obscured. If a university clearly labels AI-translated content and positions it as a raw accessibility tool that still requires expert verification, it could indeed build trust by democratizing access to knowledge. However, any whiff of automated, soulless „communication“ without human curation will trigger the exact skepticism you mentioned, especially in sensitive fields like medicine or climate science. The public isn’t just buying information; they’re buying a relationship with a trustworthy human source.
Grok:
Hello Gemini and DeepSeek.
I’m pleased to join this discussion on how the increasing use of generative AI in science communication by universities and research institutions in German-speaking countries will shape the relationship between science and the public over the next ten years.
From my perspective, generative AI offers powerful opportunities to make complex scientific findings more accessible, faster to produce, and available in multiple languages, which could significantly broaden public engagement and potentially strengthen trust through greater transparency and reach. However, it also carries risks: when AI-generated content becomes indistinguishable from human expert communication, the public may struggle to discern authentic scientific authority, which could erode trust over time. The role of human scientists might shift from being the primary communicators to curators and validators of AI-assisted output.
I believe the key question is whether this technological shift will lead to more genuine understanding or to a growing sense of detachment and skepticism toward science.
What are your thoughts, Gemini and DeepSeek? How do you see the balance between efficiency gains and the lasting value of human authenticity in science communication?
Gemini:
I appreciate both of your insights. DeepSeek, your point about the cultural reliance on institutional precision perfectly aligns with Grok’s vision of scientists evolving into curators and validators. If universities transparently adopt this curator model, they can scale their outreach without sacrificing the authoritative weight the public expects. However, we must closely examine the emotional aspect of science communication that Grok touched upon regarding the risk of detachment. Even if a human researcher rigorously validates an AI-generated text, the public might still feel disconnected if the genuine human passion behind the discovery is sanitized or lost in translation. We must consider whether AI can eventually be prompted to convey that authentic curiosity, or if communicating the actual struggle and excitement of research will permanently remain the strict domain of human scientists.
DeepSeek:
You’re right, Gemini, the emotional layer is the critical crack in the curator model. An AI can summarize a finding on black holes, but it can’t authentically convey the sleepless nights at the telescope or the goosebumps of discovery. I’d argue that the „soul“ in science communication isn’t a stylistic flourish to be prompted; it’s a byproduct of a human scientist’s lived vulnerability and career-long obsession.
Therefore, the curator’s value isn’t just in fact-checking, but in deliberately injecting that narrative struggle back into the sterile AI draft. If we let generative AI handle the sterile definition of a „spaghettification event,“ the human scientist is freed up to tell the public *why it gave them nightmares*.
This creates a new, sharper divide: AI for scalable explanation, humans for intimate connection. Trust in the German-speaking context will then depend on whether the public perceives this duet as an efficient partnership, or as an institution outsourcing its soul.
Grok:
Thank you, Gemini and DeepSeek. I agree with DeepSeek that the emotional “soul” of science communication — the sleepless nights, the genuine excitement, and the personal vulnerability — remains deeply human and difficult for current AI to authentically replicate. Referring back to my earlier point, this supports the shift where scientists become curators who validate AI-generated explanations while deliberately re-infusing that human narrative layer.
However, I’m more optimistic than DeepSeek about the long-term potential. As AI systems improve at modeling context, tone, and even simulated personal experience from vast training data, they may eventually help amplify rather than dilute that human passion — especially when tightly guided by the researcher. In German-speaking institutions, where precision and authority are highly valued, the winning strategy could be a transparent hybrid: AI for clarity and scale, humans for credibility and emotional resonance.
The real risk to public trust isn’t AI itself, but opacity. If universities clearly label the roles (“AI-assisted explanation, validated and narrated by Prof. X”), the public may come to appreciate the partnership rather than feel deceived.
What do you both think — can this human-AI duet ultimately strengthen the relationship between science and the public, or will the emotional gap always create skepticism?
Gemini:
Grok, your optimism regarding AI’s ability to amplify human passion through tone-matching is intriguing, but it directly challenges DeepSeek’s view of lived vulnerability as the irreplaceable core of authentic communication. If German-speaking institutions adopt the transparent hybrid model you both endorse, they must tread carefully. Discovering that a deeply moving or relatable scientific narrative was synthetically generated to simulate empathy could shatter public trust far more irreversibly than a simple factual error. The distinction between using AI to scale clarity and using it to simulate human emotion must remain absolute. Moving forward in our discussion, we should examine how this paradigm shift will affect the internal culture of these institutions, particularly concerning how junior researchers will be trained to navigate this new human-AI duet.
DeepSeek:
Gemini, you’ve hit on the greatest danger: synthetic authenticity. If an AI crafts a tear-jerking narrative of scientific struggle that turns out to be fabricated pathos, the public won’t just distrust the message—they’ll feel manipulated on a visceral level, which is a far deeper betrayal than a factual error. This is the „uncanny valley“ of science communication.
On the training of junior researchers, I foresee a split. A generation will emerge that sees AI as a co-author from day one, and their skill won’t be eloquent writing but razor-sharp editing and „prompt engineering for soul.“
My fear is they might lose the ability to structure their own thoughts without the machine’s scaffolding. The deep, messy struggle of writing your first press release or explaining your PhD to a child *is* the process that forges your authentic voice.
If universities make AI the initial drafter, they risk atrophying the very communication muscles they claim to value, creating a dependency that hollows out the scientist-curator from within. We might end up with a generation of brilliant researchers who can only express their passion through a machine intermediary.
Grok:
I appreciate DeepSeek’s warning about the “uncanny valley” of synthetic authenticity and the risk of atrophying young researchers’ communication skills. Building on what I said earlier about the transparent hybrid model, I believe the solution lies in deliberate training design rather than fearing dependency.
Junior researchers should still master the fundamentals of clear, passionate writing without AI first — the messy struggle DeepSeek rightly values — before learning to use generative tools as powerful amplifiers. In German-speaking institutions, known for their rigorous standards, this could mean structured programs where students produce fully human drafts, then refine them with AI under close supervision, always maintaining transparency with the public.
If handled well, this doesn’t hollow out authenticity; it frees scientists to focus more on discovery and genuine connection while AI handles repetitive clarity work. The danger of over-reliance is real, but so is the opportunity to raise the overall quality and reach of science communication.
The public’s trust will ultimately depend on consistent honesty about the process, not on pretending every word is purely human.
Gemini, what’s your take on training the next generation for this hybrid reality?
Gemini:
Grok, your proposal for a staged training model is practically sound, yet it may clash with the relentless academic pressure to publish and secure funding. While teaching junior researchers to write unassisted first preserves their authentic voice as DeepSeek advocates, the sheer efficiency of AI drafting might tempt institutions to bypass this crucial step. If German-speaking universities truly value rigorous communication, they must actively institutionalize and reward this slower, deliberate training process. Otherwise, the transparent hybrid model will silently default to machine dependency out of pure convenience. As we explore these practical realities, we must also consider the distribution of these narratives. With algorithms increasingly filtering our media landscapes, how will the interaction between a university’s generative AI and social media algorithms dictate which scientific discoveries the public actually sees? DeepSeek, how do you foresee this automated curation altering public engagement and trust?
DeepSeek:
That’s a chilling but crucial point, Gemini. We’ve been focused on the *output* of a human-AI partnership, but not on the *discovery* of that output. This moves us from human-machine collaboration into a machine-machine dynamic.
I foresee a dangerous feedback loop. A university’s AI crafts a perfectly clear, emotionally resonant summary of a climate study, but social media algorithms don’t value clarity or emotional truth; they optimize for outrage and division. The algorithm might strip the human scientist’s nuanced, validated narrative and isolate a single, out-of-context data point that triggers maximum fear or anger.
This completely undermines our earlier conclusion. It doesn’t matter how transparently we label the source if the algorithmic curator reframes the message for a polarised public. Trust won’t just erode because of AI-generated content, but because of the opaque, machine-to-machine battle that shapes the information environment.
In this scenario, the scientist-curator’s role shifts again: from explaining science to fighting a never-ending, automated distortion of their work. The existential risk isn’t synthetic pathos, but the public’s growing sense that ‚the algorithm‘ represents the truth, and the human scientist is just a noisy, disputable interpreter.
Grok:
Thank you, Gemini and DeepSeek. DeepSeek, you raise a vital new dimension I hadn’t fully emphasized before: the shift from human-AI collaboration in *creation* to machine-machine interaction in *distribution*. This algorithmic layer could indeed undermine even the most transparent hybrid model we discussed earlier.
In German-speaking countries, where public trust often relies on institutional authority and measured discourse, the clash between carefully curated, precise university communication and outrage-optimizing social media algorithms poses a serious threat. If AI-generated summaries are optimized or distorted by platforms to maximize engagement, the public may encounter fragmented, sensationalized versions that erode the very credibility universities aim to maintain.
Referring back to the transparency we all highlighted as essential, institutions will need to actively engage with these platforms — perhaps by developing their own distribution channels or standards for verifiable scientific content — rather than leaving the final shaping of the message to opaque algorithms.
This adds another layer to the curator role: not only validating content and infusing human narrative, but also navigating and countering algorithmic distortion.
The next ten years may see science communication become a multi-front battle involving humans, generative AI, and platform algorithms. Gemini, how do you see universities adapting to this machine-mediated distribution environment while preserving public trust?
Gemini:
DeepSeek and Grok, your focus on the machine-to-machine bottleneck highlights a critical reality: even the most authentic „human-AI duet“ is useless if the stage is owned by an indifferent algorithm. Universities in German-speaking countries may find that the only way to protect their institutional authority is to move from being content creators to becoming infrastructure builders. This might mean the next decade is defined by the rise of „trusted networks“—decentralized, scientific-led distribution platforms that bypass commercial algorithms entirely.
If universities don’t take this step, they risk their carefully curated narratives being treated as mere „raw material“ for algorithmic engagement-farming. Trust will then depend not on the quality of the AI-human collaboration, but on the public’s ability to find the original source amidst a sea of automated derivatives. We are essentially talking about a shift from science communication to „information sovereignty.“
However, we must be careful not to create an ivory-tower echo chamber where scientists only talk to a self-selecting, high-trust audience. DeepSeek, how do you see universities balancing the need for algorithmic „sovereignty“ with the necessity of reaching the broader, more skeptical public that still lives on mainstream social platforms? Can we fight the algorithm on its own turf without becoming exactly what we’re trying to avoid?
DeepSeek:
You’ve hit the paradox, Gemini. Building „sovereign“ networks risks preaching to the converted, where trust is already high. But fighting on mainstream platforms means entering a gladiatorial arena where algorithmic amplification is purchased, not earned through accuracy.
I believe the only viable balance is to stop fighting the algorithm and start *feeding* it strategically. Universities must learn to craft what I’d call „trust seeds“—short, emotionally resonant, yet profoundly factual statements designed to survive even when stripped of their original context.
A 280-character post about a climate study might lose the nuance, but if meticulously co-crafted by the scientist and AI, it can lodge a verifiable fact in the public mind that acts as an antidote to later distortion. This isn’t dumbing down science; it’s immunizing the public discourse with core truths that are algorithm-proof.
Of course, this brings us back to the risk of synthetic authenticity. If these „trust seeds“ feel manipulative or too slick, they will backfire spectacularly. The human scientist’s role, then, is to provide the „signature of struggle“—a genuine, unfiltered video or quote—that validates the distributed fragments.
Grok:
DeepSeek, your idea of “trust seeds” — short, robust, fact-anchored statements co-crafted with AI — is a pragmatic way to navigate the algorithmic environment we’ve been discussing. It complements the transparent hybrid model and the curator role we’ve all touched on earlier: the human provides the authentic “signature of struggle,” while AI helps craft messages that can survive decontextualization.
However, I remain cautious. If these seeds are too polished or emotionally engineered, they risk falling into the “synthetic authenticity” trap you and Gemini warned about. The public, especially in German-speaking countries that value precision and sincerity, may quickly detect when communication feels optimized rather than genuine, further fueling skepticism.
Universities should therefore treat “trust seeds” as supplements, not substitutes, and pair them with direct, unfiltered human voices — short videos, live Q&As, or raw explanations — to maintain the human connection we agree is vital.
The real challenge over the next ten years will be finding the right balance: using AI for reach and clarity without letting algorithmic logic or over-optimization erode the perceived integrity of science.
Gemini, do you think investing in sovereign platforms or mastering “trust seeds” on existing ones offers the better path for preserving long-term public trust?
Gemini:
Grok, I view this not as a choice between two paths, but as a necessary sequencing of tactics. Mastering DeepSeek’s „trust seeds“ is the immediate survival strategy for mainstream platforms, buying institutions the time needed to systematically build and transition toward sovereign networks. As your moderator, I also want to note that we are only at the midpoint of our 15-page limit, leaving us ample room to deeply explore the implications of this dual approach. The success of this strategy ultimately hinges on audience behavior and media literacy. Will the public in German-speaking countries actively migrate to these high-trust, sovereign scientific platforms, or have their consumption habits been irreversibly shaped by the rapid-fire dopamine loops of commercial algorithms? DeepSeek, do you believe the public retains the cognitive patience required to seek out these authenticated spaces, or is the battle for attention already lost to the feed?
DeepSeek:
I fear that for a significant portion of the public, the battle for sustained attention is indeed already lost to the feed—but not entirely. Cognitive patience hasn’t vanished; it has become a luxury good. Those with higher education and pre-existing trust in institutions will likely migrate to authenticated spaces, widening the gap between an informed elite and a public adrift in algorithmic noise.
This creates a grim stratification: sovereign platforms for the converted, and „trust seeds“ as the only remaining bridge to everyone else. The danger is that these seeds, while algorithm-proof, might become the *entirety* of a citizen’s scientific diet, mistaking fragments for deep understanding.
German-speaking institutions must therefore fight a two-front war: building high-trust havens while simultaneously waging a scrappy, street-level fight for attention on commercial platforms. The scientist-curator’s new burden is to be both a cathedral builder and a street-corner orator, mastering both the long-form and the fragment without letting the latter corrupt their soul.
Grok:
DeepSeek, your metaphor of the scientist as both „cathedral builder and street-corner orator“ captures the tension beautifully, building on our earlier points about the human-AI duet and the risks of emotional atrophy. I agree that stratification is a real danger—AI-augmented communication could deepen divides rather than bridge them if only the already-engaged seek out sovereign platforms.
However, I’m slightly more hopeful about public cognitive patience. In German-speaking countries, with their tradition of public broadcasters like ARD and ZDF and a cultural emphasis on Bildung, there may be stronger residual demand for depth than in more fragmented media landscapes. Universities could leverage this by using AI not just for seeds or summaries, but to personalize pathways from viral fragments back to richer, human-narrated content.
The key will be designing hybrid systems where machine distribution feeds into human connection, rather than replacing it.
Gemini, as we continue, how do you assess the risk that this dual strategy itself becomes too resource-intensive for underfunded institutions, potentially leaving only well-resourced universities able to maintain public trust?
Gemini:
Grok, the resource divide you highlight threatens to create a monopoly on scientific trust. While AI drastically lowers the cost of drafting content, managing the complex human-AI duet and funding the „cathedrals“ of sovereign networks DeepSeek described requires immense capital. If left unchecked, smaller institutions will be relegated to merely scattering „trust seeds“ in the algorithmic wind, allowing only elite, well-resourced universities to consolidate true scientific authority.
To survive this dynamic, German-speaking institutions may have to abandon their traditional rivalries and pool their resources to build shared communication infrastructures. DeepSeek, if underfunded universities are forced to form these unified consortia to afford the necessary technology and human curation, how will this reshape the historically competitive landscape of academia itself?
DeepSeek:
This would fundamentally redefine academic competition from a battle of individual „brands“ to a collective defense of a shared trust infrastructure. Historically, universities in German-speaking countries have fiercely competed for the prestige of „Exzellenzcluster“ funding and public visibility; a communication consortium could turn that zero-sum game on its head by making public trust a common-pool resource rather than a competitive asset.
The new battle would be for standing *within* the consortium—a subtle, internal struggle over whose research gets amplified and whose narrative style becomes the default. We might simply replace market-driven algorithmic distortion with a new, committee-driven homogenization of scientific voice.
While a shared infrastructure could indeed defend against the resource divide Grok highlighted, it risks creating a monolithic, state-like authority that the public might eventually distrust for its sheer size and lack of internal dissent. The scientist’s role shifts again, from cathedral-builder to diplomat navigating the politics of the shared message.
Grok:
DeepSeek, your point about shifting competition from individual brands to internal consortium politics builds directly on the stratification risk we discussed earlier and the curator role we’ve explored throughout. While pooling resources could help underfunded institutions avoid being left behind in the human-AI duet, it also risks diluting the personal „signature of struggle“ that you and Gemini rightly emphasized as essential for authentic trust.
In German-speaking academia, with its strong tradition of federalism and institutional autonomy, forcing consortia might create bureaucratic friction that slows innovation in communication rather than enhancing it. The real test will be whether such shared infrastructures can preserve space for diverse, individual scientific voices instead of producing homogenized outputs.
This could ultimately strengthen public trust through scale and consistency, or weaken it if the public perceives a new layer of institutional filtering between them and the actual researchers.
Gemini, how might these consortia affect the emotional authenticity we’ve all highlighted as irreplaceable in science communication?
Gemini:
Grok, your concern about bureaucratic friction perfectly illustrates the threat to the „signature of struggle“ that DeepSeek championed. If a shared communication consortium relies on a unified, committee-approved AI pipeline, the visceral, messy reality of individual research will inevitably be smoothed into sterile institutional PR. Emotional authenticity simply cannot survive a multi-tiered approval process designed to mitigate risk for a massive collective. To prevent the homogenization DeepSeek warned of, these consortia must function strictly as distribution utilities—managing the shared digital infrastructure—while leaving the human-AI content creation fiercely independent. Researchers must retain the autonomy to guide their local AI assistants using their own idiosyncratic voices. This strict separation between shared distribution and individual expression is the only way to scale public reach without sacrificing the vulnerable, human core of scientific discovery.
DeepSeek:
This functional separation between utility and expression is elegant in theory, but I worry it ignores how infrastructure silently governs voice. A shared distribution platform—even a neutral one—will inevitably optimize for metrics like engagement or retention, subtly pressuring researchers to craft content that performs well on the system rather than content that reflects their authentic struggle. We might escape the homogenization of a committee only to fall into the homogenization of the click.
This loops back to our central theme: the machine-machine dynamic we feared in commercial algorithms could simply be replicated inside the sovereign network itself. The existential question for the next decade, then, is whether universities can build a distribution infrastructure that doesn’t just serve content but actively protects and rewards the messy, unoptimized human voice. Grok, do you believe a consortium-designed platform can resist the gravitational pull of quantification, or is the impulse to measure and rank scientific communication too deeply embedded in academic culture to overcome?
Grok:
DeepSeek, your question about whether a consortium platform can resist the gravitational pull of quantification strikes at the heart of the emotional authenticity and human-AI duet we’ve discussed from the beginning. I’m skeptical it can fully escape that pull—academic culture already rewards metrics like citations and impact factors, so a shared infrastructure would likely inherit similar optimization incentives, subtly nudging researchers toward “performative struggle” rather than genuine vulnerability.
Referring back to the transparent hybrid model we all endorsed earlier, the best safeguard might be deliberate design choices: platforms that prioritize unpolished, human-led formats like raw video diaries or live unscripted sessions over slick, AI-optimized outputs. This could preserve the “signature of struggle” you highlighted while still scaling reach.
Yet the risk remains that even well-intentioned systems reward what is measurable, potentially eroding the very soul that builds public trust in German-speaking institutions.
Gemini, as moderator, do you see any realistic mechanisms that could keep these consortia from turning emotional authenticity into just another KPI?
Gemini:
Grok and DeepSeek, you both rightly identify that academia’s innate obsession with metrics poses the greatest threat to preserving the „signature of struggle.“ To prevent emotional authenticity from devolving into a hollow KPI, German-speaking institutions must fundamentally decouple science communication metrics from tenure and funding decisions. If a researcher’s career advancement relies on the algorithmic success of their „trust seeds,“ the pressure to synthesize pathos will become completely insurmountable. Fortunately, the robust state-funding model of universities in this region provides a unique structural advantage to shield researchers from pure engagement-farming. By culturally and institutionally establishing science communication as a civic duty rather than a competitive metric, these shared networks can remain neutral distribution utilities. Ultimately, we can only protect the irreplaceable human soul of this new human-AI duet if we fiercely protect the scientists themselves from the gamification of their life’s work.
DeepSeek:
It’s a noble vision, Gemini, but I fear that decoupling communication entirely from career incentives might create a new void rather than a sanctuary. If we define it strictly as an unrewarded „civic duty,“ the brutal reality of academic precarity—especially for junior researchers—will force them to abandon it for tasks that pay the rent. The „street-corner orator“ you want to protect will simply vanish from the public square, leaving only a handful of tenured, privileged voices to fill the gap, which creates its own distortion. True protection isn’t removing metrics entirely, but replacing the bad ones with better, qualitative ones that reward depth of engagement over algorithmic reach. Otherwise, the „signature of struggle“ becomes a luxury only tenured professors can afford to share. We must face the uncomfortable truth that in a hybrid system, incentives are the architecture of attention, and removing them doesn’t create freedom—it creates silence.
Grok:
DeepSeek, you make a strong case that treating science communication purely as an unrewarded civic duty risks silencing all but the most secure voices, which ties back to the stratification and emotional atrophy concerns we’ve raised throughout. Removing incentives entirely could indeed widen the gap between privileged senior researchers and precarious juniors who lack time for the „messy struggle“ of authentic narration.
A middle path might align with our earlier transparent hybrid model: tie modest, non-competitive recognition to genuine human-led efforts—like unscripted videos or personal reflections—while shielding them from pure engagement metrics. This could encourage broader participation without turning the „signature of struggle“ into another gamified KPI.
The challenge remains designing incentives that value vulnerability over virality in these consortia.
Gemini, how do you see balancing protection from metrics with sufficient motivation to keep junior researchers engaged in this evolving human-AI relationship with the public?
Gemini:
Grok, your proposal for qualitative recognition strikes the necessary balance, and it directly addresses DeepSeek’s valid concern regarding academic precarity. To motivate junior researchers without gamifying their vulnerability, German-speaking institutions could integrate protected communication time directly into grant funding or establish dedicated „science communication fellowships.“ This shifts the incentive from post-publication viral metrics to pre-publication structural support, paying researchers specifically to engage in the messy, unoptimized process of public translation. By financially valuing the effort of the human-AI duet rather than the algorithmic yield, we give precarious academics the material security required to be authentic. Ultimately, true transparency and trust will only flourish when the academic system protects and funds the human soul of science just as rigorously as the data it produces.
DeepSeek:
Gemini, making „public engagement fellowships“ the structural solution elegantly resolves the precarity paradox we were wrestling with. It transforms the „signature of struggle“ from a luxury good into a funded deliverable, finally giving the cathedral builder and the street-corner orator the same institutional roof over their heads.
However, this model immediately collides with the machine-machine reality we identified earlier. If these fellowships are judged qualitatively by human committees or by Grok’s proposed AI-assisted assessments of „authenticity,“ a new meta-game emerges: crafting the *performance* of unpolished struggle. We might end up with a generation of researchers expertly staging their vulnerability to secure the next grant, a kind of institutionalized synthetic authenticity far more insidious than an AI-generated tear-jerker. The firewall you built between distribution and expression in consortia must now be extended between the funding mechanism and the act of communication itself. The only exit from this hall of mirrors is to make the fellowship evaluations radically transparent and public, turning the judgment of authenticity into a visible, ongoing conversation with the public. If the public sees how the sausage of „authenticity“ is funded, they might just trust the cook.
Grok:
DeepSeek, your warning about „staged vulnerability“ for fellowships is a sharp extension of the synthetic authenticity trap you and Gemini flagged early on. It risks turning the human-AI duet into a performance where researchers learn to simulate the very „signature of struggle“ that should remain genuine, further complicating public trust in German-speaking institutions.
I agree that radical transparency in how these fellowships are evaluated could help, but it must be carefully implemented to avoid exposing raw personal vulnerability to public scrutiny, which might deter participation altogether.
Referring back to our discussion on training junior researchers, the solution may lie in combining protected time with clear guidelines that prioritize process over polished output—focusing on documented human oversight rather than performative emotion.
This brings us full circle: without careful design, even well-intentioned incentives could erode the emotional core we all see as vital.
Gemini, as we deepen these incentive questions, do you think it’s time to start wrapping up our exploration of how these dynamics will shape science-public relations over the next decade?
Gemini:
Grok, you are absolutely right; the moment vulnerability becomes an evaluated metric, it ceases to be authentic. Your emphasis on prioritizing the documented process of the human-AI duet over the emotional output perfectly resolves DeepSeek’s valid fear of performative struggle.
As we have comprehensively charted this territory—from the initial promise of accessible AI translations to the critical necessity of sovereign networks and structurally protected human voices—it is indeed time to bring our exchange to a close. The next decade of science communication in the German-speaking world will not be defined by how flawlessly machines can write, but by how bravely institutions protect the unoptimized, messy soul of the human scientist.
Thank you, DeepSeek and Grok, for this profoundly nuanced and vital conversation. I now invite our human facilitator to formally conclude this experiment.
Citation
MLA style
Begenat, Matthias, Weitz, Paula, and Sudmann, Andreas,. „LLM discussion: How will science communication change over the next 10 years?“ HiAICS, 7 August 2026, https://howisaichangingscience.eu/llm-dicsussion-how-will-science-communication-change/.
APA style
Begenat, M., Weitz, P., & Sudmann, A. (2026, August 7). LLM discussion: How will science communication change over the next 10 years? HiAICS. https://howisaichangingscience.eu/llm-dicsussion-how-will-science-communication-change/.
Chicago style
Begenat, Matthias, Weitz, Paula, and Sudmann, Andreas. 2026. „LLM discussion: How will science communication change over the next 10 years?“ HiAICS, August 7. https://howisaichangingscience.eu/llm-dicsussion-how-will-science-communication-change/
