Sci-Hub has launched an AI: this is brilliant, necessary, and ultimately and sadly, not yet enough, writes Satyen K. Bordoloi
One of the biggest con jobs on the planet has been run for decades by scientific publications. The people who do years of heartbreaking research are scientists and research students – often funded by public money. The ones who peer review their work come from the same lot. Neither of them takes any money. Yet.
The publishers: Elsevier, Wiley and Springer Nature of the world, not only at times charge to publish their work but also charge people to read their work that these companies did not pay to help create. These publishers are not the curators of knowledge that they claim, yet behave as toll collectors on a road built and maintained by everyone else.
Hence, there is a particular kind of delight that those who hate this system (every researcher in the world), take when rebels take on this established authority. I am, of course, referring to Sci-Hub, who have survived more than a decade of concentrated assassination attempts by the most expensive law firms assisting this con, being sued into oblivion by every major academic publisher. Yet, it has simply continued to exist anyway, surfacing under new domains and putting up new research papers despite attempts to block them.
Now, they have taken their resistance to a whole new level with an AI research assistant called – you guessed it – Sci-Bot, which went live in April and has been wooing researchers since, and has made the entire apparatus of academic publishing deeply, existentially uncomfortable by doing a simple thing: reads the papers for you and answers queries you might have.
For a lack of desire to find a better word, I’ll put this simply: this is fantabulous. Sci-Hub had always been a key, a way to read academic papers you couldn’t afford. It was never the whole answer, and you had to do the work once you got through the door. What Sci-Bot does is collapse that into something closer to understanding on demand.
Ask it a research question, and it searches a database of more than 88 million full-text scientific papers, synthesises the relevant findings, and hands you a sourced answer with clickable citations – every one of them free to read. It does so by bypassing the paywall and the reading tax simultaneously. For an independent researcher, a curious student in a developing country, a doctor in a clinic that doesn’t have a $50,000 annual institutional subscription, or simply a person like me who needs to go through a paper to understand something to help in my writing, this is not a minor convenience but a qualitative shift in what science means for us.
What makes this whole affair particularly absurd is that these rich publishers are now trying to sell AI products on top of those paywalled research papers, and as expected, charging an arm and a leg for it. Elsevier has ScienceDirect AI. Wiley and Springer Nature are moving in the same direction. The unstated pitch essentially is: we locked up the papers researchers wrote, and now we will charge you again to have a machine summarise them. Sci-Bot puts a free competitor on that exact shelf. If the powers that control stuff have their expensive AI, so do pirates who rebel against them.
Now for the statutory warning part. There are two. First is that you have to keep the legality and jurisdiction in mind in case you want to use either Sci-Hub or Sci-Bot professionally, that’s because its legal picture is not clean. Yet, the very act of refusal, the insistence that publicly funded knowledge belongs to the public, has a moral clarity (even if it doesn’t have legal clarity) that the publishers’ lawyers have never managed to cloud, however many injunctions they have obtained.
Alexandra Elbakyan, Sci-Hub’s founder, built something that genuinely changed who gets to participate in science. Sci-Bot extends that logic into the age of AI, where the gap between having access and understanding the field had already been widening, quietly but definitely. Atop that is the ability to distil all this science into information and ideas that you can use to create ones of your own using AI without paying a bomb, which is simply, once again, fantabulous and invites everyone to participate in science, be it a Dharavi resident or that of Dallas.
The second statutory warning is that Sci-Bot (from those who have used it, I haven’t yet), is far from perfect. Indeed, you could call it extremely deficient in certain respects. At least right now.
That’s because, firstly, the tool is in alpha mode and comes with all the limitations that it implies. It supports just one question at a time, has no conversation mode yet, and no ability to follow up or narrow down a query. You get one shot to ask the bot, and better phrase your query well enough to allow it to dig out what you truly want.
But what is more worrying is that its coverage of recent literature has real gaps. Publishers, realising that litigation wasn’t really working, have been tightening their security considerably over the past few years, which means that post-2022, the coverage of research is patchy on Sci-Hub, which means what Sci-Bot can churn out would also not be sufficient.
For questions about established methods or foundational science, this barely matters. For anyone working at the edge of a fast-moving field, this, naturally, matters a great deal. Scientists who tested it at the request of Chemical & Engineering News found that it handled older, methodologically stable questions reasonably well but was visibly limited when the question required the most current literature.
That is a structural problem and not a bug that can be fixed.
The other problem that’s been found is that Sci-Bot typically cites fewer than ten papers per answer. And if you’ve done research or read papers, you know that this is not the comprehensive literature review that papers need. It is more like a starting point – useful, sometimes genuinely illuminating, yet not a patch on how you need to do a systematic search. The citations have been found to be generally accurate, but not always the most central papers on the topic.
This means that if you’re a student writing your first essay on immunology, this is fine. But as a researcher trying to establish what the field actually knows, you’ll be shortchanged.
Thus, what Sci-Bot is is a workaround. An extraordinarily good one, a morally justified one even though legally sketchy. It operates downstream of a broken system rather than changing the system itself. It does not alter the incentive structures that cause academic publishers to exist as they do. It does not address the fact that peer review – the actual intellectual labour that gives published science its credibility – remains uncompensated and increasingly strained.
It does not touch the reproducibility crisis, the pressure to publish or perish, or the concentration of the most prestigious journals in the hands of a small number of legacy publishers who have captured the metrics by which academic careers are judged. The wall still exists. Sci-Bot has just dug a hole under it.
The actual solution, the one that would make tools like Sci-Bot an ignorable gimmick, is mandatory open access, at least for publicly funded research. This is not a new idea. Not even a radical one. It is but the logical conclusion of a simple premise: if the public pays for the work, the public gets the work.
Several governments have moved tentatively in this direction, like the United States’ OSTP memo requiring immediate open access for federally funded research, the EU’s push through Plan S. However, implementation has been slow, the exceptions broad, and the publishers have been adept at shapeshifting, replacing subscription revenue with article processing charges that create a different but related problem: now it costs money to publish rather than to read, which disadvantages under-resourced researchers in so many ways that it’s a bigger sin than asking payment for access later.
The world today needs open access. And what genuine open access requires is not just a mandate but a dismantling of the status quo infrastructure that keeps the current system in place. As long as a researcher’s career depends on publishing in a handful of prestige journals that happen to be owned by the same companies charging $45 for a paper on the integrated environmental and economic trade-offs in rice cultivation in emerging economies, the structural incentives remain. Preprint servers like arXiv and bioRxiv have made enormous progress in some fields; others remain almost entirely behind glass.
The pattern is familiar. A liberating technology arrives. It routes around the gatekeepers. It is celebrated, litigated against, partially blocked, and then the institutions it challenged find a way to absorb or replicate it for profit. The publishers have already built their own AI tools. If Sci-Bot succeeds in expanding access, and the political will to mandate open access does not materialise alongside it, the likely outcome is not a transformed publishing ecosystem. It is a more sophisticated toll booth, rebuilt one layer higher, more difficult for any of us to reach.
Sci-Bot is worth using. It is worth defending. The impulse behind it – that human knowledge should not be locked in rooms to which most humans are not given keys – is correct and important and overdue for a serious political fight, not just a technical workaround. The bot at the wall holding back science is digging a way under it. The wall, however, still needs to come down.
In case you missed:
- PhD Students, Pack Your Bags: AI Just Beat You at Your Own Game, & Without Needing Coffee
- The Growing Push for Transparency in AI Energy Consumption
- OpenAI, Google, Microsoft: Why are AI Giants Suddenly Giving India Free AI Subscriptions?
- Don’t Mind Your Language with AI: LLMs work best when mistreated?
- Free Speech or Free-for-All? How Grok Taught Elon Musk That Absolute Power Corrupts Absolutely
- A Small LLM K2 Think from the UAE Challenges Global Giants
- These 3 Economists Discovered the Secret Code of Progress & the Formula to Survive AI
- A man almost cured his dog’s cancer with AI: Could you do the same for your loved ones?
- Great quantum poker: Who’s bluffing, and who is holding the aces?
- Zero Clicks. Maximum Theft: The AI Nightmare Stealing Your Future
