What should we make of OpenAI’s GPT-4, anyway? Is the large language model a major step on the way to an artificial general intelligence (AGI)—the insider’s term for an AI system with a flexible human-level intellect? And if we do create an AGI, might it be so different from human intelligence that it doesn’t see the point of keeping Homo sapiensaround?

If you query the world’s best minds on basic questions like these, you won’t get anything like a consensus. Consider the question of GPT-4’s implications for the creation of an AGI. Among AI specialists, convictions range from Eliezer Yudkowsky’s view that GPT-4 is a clear sign of the imminence of AGI, to Rodney Brooks’s assertion that we’re absolutely no closer to an AGI than we were 30 years ago.

On the topic of the potential of GPT-4 and its successors to wreak civilizational havoc, there’s similar disunity. One of the earliest doomsayers was Nick Bostrom; long before GPT-4, he argued that once an AGI far exceeds our capabilities, it will likely find ways to escape the digital world andmethodically destroy human civilization. On the other end are people like Yann LeCun, who reject such scenarios as sci-fi twaddle.

In between are researchers who worry about the abilities of GPT-4 and future instances of generative AI to cause major disruptions in employment, to exacerbate the biases in today’s society, and to generate propaganda, misinformation, and deep fakery on a massive scale. Worrisome? Yes, extremely so. Apocalyptic? No.

Many worried AI experts signed an open letter in March asking all AI labs to immediately pause “giant AI experiments” for six months. While the letter didn’t succeed in pausing anything, it did catch the attention of the general public, and suddenly made AI safety a water-cooler conversation. Then, at the end of May, an overlapping set of experts—academics and executives—signed a one-sentence statementurging the world to take seriously the risk of “extinction from AI.”

Below, we’ve put together a kind of scorecard. IEEE Spectrum has distilled the published thoughts and pronouncements of 22 AI luminaries on large language models, the likelihood of an AGI, and the risk of civilizational havoc. We scoured news articles, social media feeds, and books to find public statements by these experts, then used our best judgment to summarize their beliefs and to assign them yes/no/maybe positions below. If you’re one of the luminaries and you’re annoyed because we got something wrong about your perspective, please let us know. We’ll fix it.

And if we’ve left out your favorite AI pundit, our apologies. Let us know in the comments section below whom we should have included, and why. And feel free to add your own pronouncements, too.

. 

SAM ALTMAN

CEO, OpenAI

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Maybe

Core belief about the future of large language models

LLMs are an important step toward artificial general intelligence, but it will be a “slow takeoff” in which we’ll have time to address the real risks that the technology brings.

Quote

“Future versions of AI will solve some of our most pressing problems, really increase the standard of life, and also figure out much better uses for human will and creativity.”

Photo: OpenAI

JACOB ANDREAS

Professor, MIT

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Large language models such as GPT-4 offer “a massive space of possibilities,” but even the best models today are not reliable or trustworthy, and a lot of work will be required to fix these problems.

Quote

“The thing that I’m most scared about has to do with… truthfulness and coherence issues.”

Photo: Gretchen Ertl/MIT

EMILY M. BENDER

Professor, University of Washington

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Powerful corporations’ hype about LLMs and their progress toward AGI create a narrative that distracts from the need for regulations that protect people from these corporations.

Quote

“We can imagine other futures, but to do so, we have to maintain independence from the narrative being pushed by those who believe that AGI is desirable and that LLMs are the path to it.”

Photo: Emily M. Bender

YOSHUA BENGIO

Professor, University of Montreal

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Maybe

Is an AGI likely to cause civilizational disaster if we do nothing?

Maybe

Core belief about the future of large language models

We are not prepared for the proliferation of powerful AI tools. The danger will not come from these tools becoming autonomous but rather from military applications and from misuse of the tools that sows disinformation and discrimination.

Quote

“We have passed a critical threshold: Machines can now converse with us and pretend to be human beings. This power can be misused for political purposes at the expense of democracy.”

Photo: Camille Gladu-Drouin

NICK BOSTROM

Professor, University of Oxford

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Yes

Core belief about the future of large language models

Sentience is a matter of degrees, and today’s LLMs could be considered to have some small degree of sentience. Future versions could have more.

Quote

“Variations of these AIs may soon develop a conception of self as persisting through time, reflect on desires, and socially interact and form relationships with humans.”

Photo: University of Oxford

RODNEY BROOKS

Professor emeritus, MIT, and cofounder, RobustAI

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

The rise of large language models and the accompanying furor is not much different from many previous such upheavals in technology in general and AI in particular.

Quote

“What the large language models are good at is saying what an answer should sound like, which is different from what an answer should be.”

Photo: Christopher Michel

SÉBASTIEN BUBECK

Research manager, Microsoft Research

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Maybe

Core belief about the future of large language models

GPT-4 has a type of generalized intelligence and exhibits a flexible understanding of concepts. But it lacks certain fundamental building blocks of human intelligence, such as memory and the ability to learn.

Quote

“All of the things I thought [GPT-4] wouldn’t be able to do? It was certainly able to do many of them—if not most of them.”

Photo: Microsoft Research

JOY BUOLAMWINI

Founder, Algorithmic Justice League

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

While LLMs are not likely to lead to a civilization-crushing AGI, people’s fear of that hypothetical can be used to slow down the progress of AI and address pressing near-term concerns.

Quote

“Honest  question: If you believe you are unleashing the end of the world, why continue your current path?”

Photo: Getty Images

TIMNIT GEBR

Founder, Distributed AI Research Institute

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Discussion of AGI is a deliberate distraction from the real and pressing risks of today’s LLMs. The companies responsible need regulations to increase transparency and accountability and to end exploitative labor practices.

Quote

“Congress needs to focus on regulating corporations and their practices, rather than playing into their hype of ‘powerful digital minds.’”

Photo: Timnit Gebru

ALISON GOPNIK

Professor, UC Berkeley

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Human intelligence evolved from our interactions with the natural world. It’s unlikely that LLMs will become intelligent through text alone.

Quote

“What [LLMs] let us do is take all the words, all the text that people have written over all time, and summarize those in a way that is effective and lets us interact. I think what it isn’t is a new kind of intelligence.”

Photo: Gary Doak/Alamy

DAN HENDRYCKS

Director and cofounder, Center for AI Safety

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Yes

Core belief about the future of large language models

Economic and strategic pressures are pushing AI companies to develop powerful AI systems that can’t be reliably controlled. Unless AI safety research becomes a worldwide priority, humanity is in danger of extinction.

Quote

“Whereas AI researchers once spoke of ‘designing’ AIs, they now speak of ‘steering’ them. And even our ability to steer is slipping out of our grasp as we let AIs teach themselves and increasingly act in ways that even their creators do not fully understand.”

Photo: Dan Hendrycks

GEOFFREY HINTON

Professor, University of Toronto

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Maybe

Core belief about the future of large language models

LLM technology is advancing too rapidly; it shouldn’t advance any further until scientists are confident that they can control it.

Quote

“Until quite recently, I thought it was going to be like 20 to 50 years before we have general-purpose AI. And now I think it may be 20 years or less.”

Photo: University of Toronto

CHRISTOF KOCH

Chief scientist, MindScope Program, Allen Institute

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Maybe

Core belief about the future of large language models

GPT-4 and other large language models will exceed human capabilities in some endeavors, and approach AGI capabilities, without human-type understanding or consciousness.

Quote

“What it shows, very clearly, is that there are different routes to intelligence.“

Photo: Allen Institute

JARON LANIER

Computer scientist, entrepreneur, author, artist

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Generative AI and chatbots are an important advance because they offer users choice. By offering a range of responses, these models will erode the illusion of the “monolithic truth” of the Internet and AI.

Quote

“This idea of surpassing human ability is silly because it’s made of human abilities. It’s like saying a car can go faster than a human runner. Of course it can, and yet we don’t say that the car has become a better runner.”

Photo: Jaron Lanier

YANN LECUN

Chief AI scientist, Meta

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

GPT-4 itself is dazzling in some ways, but it is not particularly innovative. And GPT-style large language models will never lead to an AI with common sense and real, human-type understanding.

Quote

“The amplification of human intelligence by machine will enable a new renaissance or a new age of enlightenment, propelled by an acceleration of scientific, technical, medical, and social progress thanks to AI.”

Photo: Meta

GARY MARCUS

Professor, NYU

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

Maybe

Core belief about the future of large language models

Making large language models more powerful is necessary but not sufficient to create an AGI.

Quote

“There is still an immense amount of work to be done in making machines that truly can comprehend and reason about the world around them.”

Photo: NYU

MARGARET MITCHELL

Chief ethics scientist, Hugging Face

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Narratives that present AGI as ushering in either a utopia or extinction both contribute to hype and disguise the concentration of power in a handful of corporations.

Quote

“Ignoring active harms right now is a privilege that some of us don’t have.”

Photo: Hugging Face

MELANIE MITCHELL

Professor, Sante Fe Institute

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

Both the abilities and the dangers of LLMs are overhyped. We should be thinking more about LLMs’ immediate risks, such as disinformation and displays of harmful bias.

Quote

“We humans are continually at risk of over-anthropomorphizing and over-trusting these systems, attributing agency to them when none is there.”

Photo: Melanie Mitchell

ANDREW NG

Founder and CEO, Landing AI

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

While the latest LLMs are far from being AGIs, they have some superhuman capabilities that can be harnessed for human advancement.

Quote

“In the past year, I think we’ve made one year of wildly exciting progress in what might be a 50- or 100-year journey.”

Photo: Andrew Ng

MAX TEGMARK

Professor, MIT and president, Future of Life Institute

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Yes

Core belief about the future of large language models

More powerful LLMs and AI systems should be developed only once researchers are confident that their effects will be positive and their risks will be manageable.

Quote

“Our letter mainstreamed pausing; [the statement from the Center for AI Safety] mainstreams extinction. Now a constructive open conversation can finally start.”

Photo: Max Tegmark

MEREDITH WHITTAKER

President, Signal Foundation

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

No

Is an AGI likely to cause civilizational disaster if we do nothing?

No

Core belief about the future of large language models

LLMs are already doing serious damage to society by reflecting all the biases inherent in their training data.

Quote

“Why do we need to create these? What are the collateral consequences of deploying these models in contexts where they’re going to be informing people’s decisions?”

Photo: Signal Foundation

ELIEZER YUDKOWSKY

Cofounder, Machine Intelligence Research Institute

Is the success of GPT-4 and today’s other large language models a sign that an AGI is likely?

Yes

Is an AGI likely to cause civilizational disaster if we do nothing?

Yes

Core belief about the future of large language models

LLMs could lead to an AGI, and the first superintelligent AGI will, by default, kill literally everyone on the planet. And we are not on track to solve this problem on the first try.

Quote

“Unaligned operation at a dangerous level of intelligence kills everybody on Earth, and then we don’t get to try again.”

Article link: https://spectrum.ieee.org/artificial-general-intelligence

July 5, 2023

Captain James A. Lovell Federal Health Care Center (FHCC) will receive the same single, common federal electronic health record (EHR) as other Department of Defense (DOD) and Department of Veterans Affairs (VA) sites. The federal EHR provides flexibility to configure the EHR to meet specific facilities’ needs while still maintaining interoperability between the Departments. Through established governance and change control processes, DOD, VA and Department of Homeland Security’s U.S. Coast Guard (USCG) sites can each request configuration changes as long as these changes do not interfere or prohibit interoperability between the Departments. As a result, any Departments using the federal EHR can access these changes and benefit from additional capabilities their facility may need.

As a fully integrated joint sharing site, Lovell FHCC does require unique approaches to some configurations and deployment activities. These include the following:

Patient Care Location (PCL) Hierarchies. PCL hierarchies correspond to physical locations of patients receiving health care services, with facilities at the top level of the hierarchy followed by buildings, nursing units, rooms and beds. Unlike other sites that use either a DOD or VA PCL hierarchy, Lovell FHCC will use two PCL hierarchies—one for each Department, in their respective facilities. Using both DOD and VA PCL hierarchies enables each Department to satisfy their respective statutory requirements.
• Provider User Role Assignments: User roles enable Department-specific workflows, and assignments determine the training staff receive on the federal EHR. Lovell FHCC staff will be assigned a DOD or VA user role based on PCL alignment of the clinic or location where they provide care, not their Department alignment. Some areas, such as pharmacy, behavioral health and occupational health, will require staff assignment of both DOD and VA roles.
• Deployment Responsibilities: The FEHRM, DoD Healthcare Management System Modernization Program Management Office and VA’s Electronic Health Record Modernization Integration Office have roles and responsibilities in deploying the federal EHR at Lovell FHCC. These responsibilities vary for each of the high-level deployment events. For the adoption and site activation deployment events, DOD will lead, working closely with VA to supplement. Conversely, each Department will provide its respective training, infrastructure and devices to Lovell FHCC, based on PCL alignment. Regardless of role alignment, the Departments will execute all deployment events in a coordinated manner.
• Patient Portals. Lovell FHCC will use DOD and VA patient portals. DOD beneficiaries will use the DOD patient portal, and VA beneficiaries will use the VA patient portal. Dual eligible patients can use either portal.

Article link: https://www.linkedin.com/pulse/unique-approaches-lovell-fhcc-federal-ehr-deployment-fehrm

Science is about to become much more exciting — and that will affect us all, argues Google’s former CEO.

By Eric Schmidt July 5, 2023

It’s yet another summer of extreme weather, with unprecedented heat waves, wildfires, and floods battering countries around the world. In response to the challenge of accurately predicting such extremes, semiconductor giant Nvidia is building an AI-powered “digital twin” for the entire planet. 

This digital twin, called Earth-2, will use predictions from FourCastNet, an AI model that uses tens of terabytes of Earth system data and can predict the next two week’s weather tens of thousands of times faster and more accurately than current weather forecasting methods.

Usual weather prediction systems have the capacity to generate around fifty predictions for the week ahead. FourCastNet can instead predict thousands of possibilities, accurately capturing the risk of rare but deadly disasters and thereby giving vulnerable populations valuable time to prepare and evacuate. 

The hoped-for revolution in climate modeling is just the beginning. With the advent of AI, science is about to become much more exciting — and in some ways unrecognizable. The reverberations of this shift will be felt far outside the lab and will affect us all. 

If we play our cards right with sensible regulation and proper support for innovative uses of AI to address science’s most pressing issues, AI can rewrite the scientific process. We can build a future where AI-powered tools will both save us from mindless and time-consuming labor and also propose creative inventions and discoveries, encouraging breakthroughs that would otherwise take decades.

Although AI in recent months has become almost synonymous with large language models, or LLMs, in science, there are a multitude of different model architectures that may drive even bigger impacts. In the past decade, most progress in science so far has been through smaller, “classical” models focused on specific questions. These models have already brought about profound advances in science. More recently, larger deep learning models that are beginning to incorporate cross-domain knowledge and generative AI have expanded what is possible.

Scientists at McMaster and MIT, for example, used an AI model to identify an antibiotic for a drug-resistant pathogen that the World Health Organization labeled as one of the world’s most dangerous antibiotic-resistant bacteria for hospital patients. A Google DeepMind model can control plasma in nuclear fusion reactions, bringing us closer to a clean energy revolution. Within healthcare, the FDA has already cleared 523 devices that use AI — 75 percent for use in radiology.

Reimagining science

At its core, the scientific process we all learned in elementary school will remain the same: conduct background research, identify a hypothesis, test it through experimentation, analyze the collected data, and reach a conclusion. But AI has the potential to revolutionize how each of these components looks in the future. 

Artificial intelligence is already transforming how some scientists conduct literature reviews. Tools like PaperQA and Elicit harness LLMs to scan databases of articles and produce succinct and accurate summaries of the existing literature — citations included.

Once the literature review is complete, scientists form a hypothesis to be tested. LLMs at their core work by predicting the next word in a sentence, thereby building up to entire sentences and paragraphs. This technique makes LLMs uniquely suited to scaled problems intrinsic to science’s hierarchical structure and can enable such models to predict the next big discovery in physics or biology.

AI can also spread the search net for hypotheses wider and narrow the net more quickly. As a result, AI tools can help formulate stronger hypotheses, such as models that spit out more promising candidates for new drugs. We’re already seeing simulations running multiple orders of magnitude faster than just a few years ago, allowing scientists to try more design options in simulation before carrying out real-world experiments. 

Scientists at CalTech, for example, used an AI fluid simulation model to automatically design a better catheter that prevents bacteria from swimming upstream and causing infections. This will fundamentally shift the incremental process of scientific discovery, allowing researchers to design for the optimal solution from the outset rather than progress through a long line of progressively better designs, as we saw in years of innovation on filaments in lightbulb design.

Moving onto the experimentation step, AI will be able to conduct experiments faster, cheaper, and at greater scale. For example, we can build AI-powered machines with hundreds of micropipettes running day and night to create samples at a rate no human could match. Instead of limiting themselves to just six experiments, scientists can use AI tools to run one thousand.

Scientists who are worried about their next grant, publication, or tenure process will no longer be bound to safe experiments with the highest odds of success, instead free to pursue bolder and more interdisciplinary hypotheses. When evaluating new molecules, for example, researchers tend to stick to candidates similar in structure to those we already know, but AI models do not have to have the same biases and constraints. 

Eventually, much of science will be conducted at “self-driving labs” — automated robotic platforms combined with artificial intelligence. Here, we can bring AI prowess from the digital realm into the physical world. Such self-driving labs are already emerging at companies like Emerald Cloud Laband Artificial and even at Argonne National Laboratory. 

Finally, at the stage of analysis and conclusion, self-driving labs will move beyond automation and, informed by experimental results they produced, use LLMs to interpret the results and recommend the next experiment to run. Then, as partners in the research process, the AI lab assistant could order supplies to replace those used in earlier experiments and set up and run the next recommended experiments overnight with results ready to deliver in the morning — all while the experimenter is home sleeping.

Possibilities and limitations

Young researchers might be shifting nervously in their seats at the prospect. Luckily, the new jobs that emerge from this revolution are likely to be more creative and less mindless than most current lab work. 

AI tools can lower the barrier to entry for new scientists and open up opportunities to those traditionally excluded from the field. With LLMs able to assist in building code, STEM students will no longer have to master obscure coding languages, opening the doors of the ivory tower to new, nontraditional talent and making it easier for scientists to engage with fields beyond their own. Soon, specifically trained LLMs might move beyond offering first drafts of written work like grant proposals and might be developed to offer “peer” reviews of new papers alongside human reviewers.

AI tools have incredible potential, but we must recognize where the human touch is still important and avoid running before we can walk. For example, successfully melding AI and robotics through self-driving labs will not be easy. There is a lot of tacit knowledge that scientists learn in labs that is difficult to pass to AI-powered robotics. Similarly, we should be cognizant of the limitations — and even hallucinations — of current LLMs before we offload much of our paperwork, research, and analysis to them. 

Companies like OpenAI and DeepMind are still leading the way in new breakthroughs, models, and research papers, but the current dominance of industry won’t last forever. DeepMind has so far excelled by focusing on well-defined problems with clear objectives and metrics. One of its most famous successes came at the Critical Assessment of Structure Prediction, a biennial competition where research teams predict a protein’s exact shape based on the order of its amino acids. 

From 2006 to 2016, the average score in the hardest category ranged from around 30 to 40 on CASP’s scale of one to 100. Suddenly, in 2018, DeepMind’s AlphaFold model scored a whopping 58. An updated version called AlphaFold2 scored 87 two years later, leaving its human competitors even further in the dust.

Thanks to open-source resources, we’re beginning to see a pattern where industry hits certain benchmarks and then academia steps in to refine the model. After DeepMind’s release of AlphaFold, Minkyung Baek and David Baker at the University of Washington released RoseTTAFold, which uses DeepMind’s framework to predict the structures of protein complexesinstead of only the single protein structures that AlphaFold could originally handle. More importantly, academics are more shielded from the competitive pressures of the market and so can venture beyond the well-defined problems and measurable successes that attract DeepMind. 

In addition to reaching new heights, AI can help verify what we already know by addressing science’s replicability crisis. Around 70 percent of scientists report having been unable to reproduce another scientist’s experiment — a disheartening percentage. As AI lowers the cost and effort of running experiments, it will in some cases be easier to replicate — or fail to replicate — results, contributing to a greater trust in science.

The key to replicability and trust is transparency. In an ideal world, everything in science would be open access, from articles without paywalls to open-source data, code, and models. Sadly with the dangers that such models are able to unleash, it isn’t always realistic to make all models open source. In many cases, the risks of being completely transparent outweigh the benefits of trust and equity. Nevertheless, to the extent that we can be transparent with models — especially classical AI models with more limited uses — we should be. 

The importance of regulation

With all these areas, it’s essential to remember the inherent limitations and risks of artificial intelligence. AI is such a powerful tool because it allows humans to accomplish more with less: less time, less education, less equipment. But these capabilities make it a dangerous weapon in the wrong hands. University of Rochester professor Andrew White was contracted by OpenAI to participate in a “red team” that could expose GPT-4’s risks before it was released. Using the language model and giving it access to tools, White found it could propose dangerous compounds and even order them from a chemical supplier. To test the process, he had a (safe) test compound shipped to his house the next week. OpenAI says it used his findings to tweak GPT-4 before it was released.

Even humans with entirely good intentions can still prompt AIs to produce bad outcomes. We should worry less about creating the Terminator and, as computer scientist Stuart Russell put it, more about becoming King Midas, who wished for everything he touched to turn to gold and thereby accidentally killed his daughter with a hug.

We have no mechanism to prompt an AI to change its goal, even when it reacts to its goal in a way we don’t anticipate. One oft-cited hypothetical asks you to imagine telling an AI to produce as many paperclips as possible. Determined to accomplish its goal, the model hijacks the electrical grid and kills any human who tries to stop it as the paperclips keep piling up. The world is left in shambles. The AI pats itself on the back; it has done its job. (In a wink to this famous thought experiment, many OpenAI employees carry around branded paperclips.)

OpenAI has managed to implement an impressive array of safeguards, but these will only remain in place as long as GPT-4 is housed on OpenAI’s servers. The day will likely soon come when someone manages to copy the model and house it on their own servers. Such frontier models need to be protected to prevent thieves from removing the AI safety guardrails so carefully added by their original developers.

To address both intentional and unintentional bad uses of AI, we need smart, well-informed regulation — on both tech giants and open-source models — that doesn’t keep us from using AI in ways it can be beneficial to science. Although tech companies have made strides in AI safety, government regulators are currently woefully underprepared to enact proper laws and should take greater steps to educate themselves on the latest developments.

Beyond regulation, governments — along with philanthropy — can support scientific projects with a high social return but little financial return or academic incentive. There are several areas with an especially urgent time to solution, including climate change, biosecurity, and pandemic preparedness. It is in these areas that we most urgently need the speed and scale that AI simulations and self-driving labs offer. 

Government can also help develop large, high-quality datasets such as those on which AlphaFold relied — insofar as safety concerns allow. Open datasets are public goods: They benefit many researchers, but researchers have little incentive to create them themselves. Government and philanthropic organizations can work with universities and companies to pinpoint seminal challenges in science that would benefit from access to powerful databases. 

Chemistry, for example, has one language that unites the field, which would seem to lend itself to easy analysis by AI models. But no one has properly aggregated data on molecular properties stored across dozens of databases, which keeps us from accessing insights into the field that would be within reach of AI models if we had a single source. Biology, meanwhile, lacks the known and calculable data that underlies physics or chemistry, with subfields like intrinsically disordered proteins still a mystery to us. It will therefore require a more concerted effort to understand — and even record — the data for an aggregated database.

The road ahead to broad AI adoption in the sciences is long, with a lot that we must get right, from building the right databases to implementing the right regulations, mitigating biases in AI algorithms to ensuring equal access across borders to resources like compute and GPUs. 

Nevertheless, this is a profoundly optimistic moment. Previous paradigm shifts in science, like the emergence of the scientific process or big data, have been inwardly focused — making science more precise, accurate, and methodical. AI meanwhile is expansive, allowing us to combine information in novel ways and to bring creativity and progress in the sciences to new heights.

BIO: Eric Schmidt was the CEO of Google from 2001 to 2011. He is currently co-Founder of Schmidt Futures, a philanthropic initiative that bets early on exceptional people making the world better, applying science and technology, and bringing people together across fields.

Article link: https://www-technologyreview-com.cdn.ampproject.org/c/s/www.technologyreview.com/2023/07/05/1075865/eric-schmidt-ai-will-transform-science/amp/