Artificial intelligence is advancing faster than its researchers and proponents can keep up. The authors of Stanford University’s latest longitudinal study of AI – conducted every five years – had a lot of catching up to do. Since the last survey of AI developments in 2021, generative AI broke onto the scene, along with countless other advancements.

While AI has become a force that promises exponentially more benefits than the risks it is posing in terms of economic opportunities and scientific discoveries, more efforts need to be made to assure AI is open and collaborative to all, according to the just-released edition of the “ One Hundred Year Study on Artificial Intelligence (AI100) .” The study was first launched in 2014.

There has been astounding progress in AI since 2021, the Stanford team states. "Challenges that appeared distant – fluent natural language interaction, broad-spectrum question answering, competent code generation – now seem largely addressed, at least in their shallow forms," the study team, led by Michael Wooldridge, professor at the University of Oxford, observes. AI has moved from theory to practical applications, such as drug discovery and autonomous transportation.

The following are some of the changes AI has wrought over the past five years, the Stanford team finds:

Creative and cognitive fields are being disrupted like never before. Creative work, software development, and even pure mathematics were seen as “too cognitively challenging for automation by AI,” the report observes. “All of these areas are now being disrupted, as generative AI proves itself increasingly capable at low-level creative work, computer programming, and mathematical proof. The genie will not go back in the bottle, and these areas will need to reinvent themselves in the years ahead, just as other areas disrupted by technology have done.”

AI power is too concentrated. The Stanford team expressed concern that frontier AI development is concentrated among a small number of US and Chinese entities, “raising international concerns about who will ultimately own and have access to leading-edge AI technologies.”

Our trust in imagery is undermined. Until recently, a photo or video was all the proof one needed that an event actually took place or that a person actually existed. Now, “Seeing is no longer believing,” the report states. “AI-generated images, audio, and video are increasingly difficult to distinguish from authentic recordings. Cloned voices are increasingly being used to deceive the public and extort money from unwitting victims, and synthetic media – deepfakes – have been used to influence elections around the world. The implications will be broadly felt – from journalism and creative industries to law, and beyond, and may contribute to the continuing erosion of public trust.”

Smaller models for AI are inevitable. For anyone working within the AI space, scaling has been the pathway to converting AI from experiments to practical reality. However, the Stanford team warns, “the scaling paradigm is now under pressure because new sources of high-volume, high-quality data are scarce and compute costs for frontier AI models are prohibitive.”

Their prediction: "Expect the next phase of development to be defined less by repeated training of ever-larger models and more by smarter algorithms and new ways of using potentially smaller models with less frequent rebuilding." This will provide models “the right context and memory, more time to reason during inference, specialized training to do so, and tools or agents to solve complex and siloed tasks.”

AI’s impact on innovation and human behavior is uncertain. The report discusses AI’s contributions to scientific discovery, and innovation in general, raising concerns about the role of humans in AI-assisted processes. “Experimental design may be able to maximize the benefits of collaboration between scientists and AI, but how and to what extent to keep humans in the loop is an important open question,” they state. "We need to better understand how AI and humans are codeveloping and how this is influencing human behavior, language, and culture. We will need to understand how AI-mediated information environments reshape collective cognition and moral behavior." Of particular concern is "how cultural innovation occurs in a world where an increasing proportion of the information people encounter has been generated or filtered by a small number of large AI systems."

Security hangs in the balance. “AI systems have demonstrated the ability both to identify serious security vulnerabilities in computer code and to successfully carry out cyberattacks via teams of AI agents,” the report cautions. “As well as being an offensive weapon for cyberattacks, AI systems will also play an important role in securing and defending society’s computer systems and computer-controlled infrastructure.”

Tellingly, “this report will be out of date by the time you read it,” the Stanford co-authors state. “We have repeatedly seen established wisdom about the capabilities of AI systems overturned by technological developments, and benchmarks intended to define a roadmap for decades become irrelevant within months.”