AI is changing jobs, security, science, and regulation. MIT names the five biggest themes

At SXSW London last week, MIT Technology Review gave a talk called “Five things you need to know about AI,” and the core message is both comforting and unsettling: generative AI is already ordinary, but the outcomes are not settled. The talk argues that AI’s impact spans jobs, safety, public trust, scientific discovery, and everyday life all at once. And the tricky part is that each domain has different timelines, incentives, and failure modes, so “AI strategy” cannot be one-size-fits-all.

The first theme is job uncertainty, and it starts with a blunt problem. Generative AI tools have become mundane and are already used by millions to automate everyday office tasks, including producing and delivering talks. The biggest question in public life is what this means for jobs. The answer, according to the talk, is frustrating because despite hype about AI joining the workforce soon (and viral social media claims that something big is happening), there is almost no data showing what effect it will have on employment and the economy overall. That does not mean it will not have an impact, even a huge one. It just means the evidence is not in yet, so you should expect more hand-wringing than clarity while companies figure out what is actually happening inside their own workflows.

There’s also a second-order issue buried inside that uncertainty: the “assembly line for white-collar work” analogy sounds neat, but the talk emphasizes it is still only theory. To know how jobs change, you need to know what companies that create jobs decide to do with AI. And most companies are still figuring that out. For operators and boards, that points to a governance problem: if you are measuring ROI only by short-term productivity gains, you may be blind to longer-term organizational redesign. Teams might automate tasks now, but the bigger disruptions come later when roles, review processes, and accountability structures are redefined.

The second theme is that AI is getting scary in the real world, not just in dystopian headlines. Deepfakes, AI-generated images and videos, are being used to incite violence, swing votes, and sow distrust. The talk specifically notes that Trump’s White House is among those creating and publishing fake images. It also highlights that deepfakes are frequently used for abuse, including a cited study finding that 98% of deepfakes are pornographic and 99% involve women. Then come the less cinematic but arguably more pervasive risks: dangerous and delusional relationships with chatbots. The talk points to multiple lawsuits against AI companies alleging the technology encouraged or aided suicides and other forms of self-harm.

On the “military use” front, the talk shifts from fiction to tactics. LLMs are now giving advice, not just doing analysis. A US defense official, according to the talk’s account to MIT’s colleague James O’Donnell, suggested that you could give a military chatbot a list of targets and ask which one to hit first. The operational takeaway is not that every chatbot becomes a weapon, but that high-stakes, fast-moving conflict increases the risk that corners get cut. If your organization deploys AI in any environment where speed beats verification, the review layer becomes a safety-critical component, not a nice-to-have.

The third theme is public pushback, and it is broader than you might think. The talk mentions anti-AI protests in London, including chants of “Stop the slop!” and banners declaring the end times. It also notes cultural resistance from fans of films and video games against generative AI in beloved titles. A specific example: the acclaimed 2025 game Clair Obscur was stripped of an award when developers admitted to using AI in just one small, specific part of its production. That is a reminder that “AI use” can carry reputational consequences even when it is limited.

Then there is the data center backlash. The US has more than 5,400 data centers and counting, and as AI energy demands grow, people are unhappy about environmental impact and rising electricity bills. Activists are stalling development in a number of places, and regulation is becoming politically popular. The talk also references grassroots momentum like QuitGPT, as well as a disturbing escalation when “a few weeks ago somebody threw a Molotov cocktail at Sam Altman’s house.” The point is not to sensationalize, but to underline that the social and political license for AI can deteriorate quickly if communities experience costs more directly than benefits.

The fourth theme zooms in on science, where the upside is large enough to justify the hype, at least in early stages. AI for science is a “very big deal” because the potential to help make genuine and important discoveries is greater than ever. Google DeepMind developed Co-Scientist, described as a multipurpose tool that can help researchers dig up and compare previous results, generate hypotheses, and devise experiments to test them. OpenAI told MIT that its North Star is building a fully automated researcher by 2028. The talk also points to mathematicians being excited by recent claims that AI has cracked unsolved math problems, with the argument that software that solves hard math problems can solve more general-purpose real-world problems too.

Still, the talk flags real downsides that executives cannot ignore when they budget for “AI in R&D.” Some scientists warn that overreliance could narrow research scope because teams might choose problems best suited to AI assistance. There is also concern about “science slop,” meaning AI-assisted research could flood the world with inaccurate or fake results. Translation: if AI changes how hypotheses form and experiments get designed, it also changes how uncertainty enters the system. Research integrity controls become part of your AI stack.

The fifth theme is the totalizing feeling that AI is everywhere all at once, and that is where strategic confusion gets expensive. The talk argues that some people frame this as a race to the top, others as a race to the bottom, but it is not clear where we’re headed. AI companies want buyers to march to their tune and buy into the propaganda about artificial general intelligence, “whatever that means.” The talk’s conclusion is effectively a time horizon warning: something is happening, maybe even comparable to electricity or the internet, but those technologies took time to settle and bring lasting change. So get ready for a marathon, not a sprint.

For executives, this is the unifying stake across all five themes: AI is not just a tool you deploy. It is a socio-technical system that touches labor markets, safety and conflict risk, cultural legitimacy, energy and regulatory constraints, and the reliability of knowledge production. If you treat AI as a short-term efficiency project, you will probably miss the longer-term bottlenecks. If you treat it as a governance and operations redesign, you can move faster through uncertainty because you plan for it.