From chatbots to superintelligence: charting AI's ambitious journey

Is humanity on the brink of intelligence superiority? Some believe we are on the brink of such a development. Last week, Ilya Sutskever announced Safe Superintelligence, Inc. (SSI), a new startup dedicated to building advanced artificial superintelligence (ASI) models, a hypothetical AI that far exceeds human capabilities. In a statement about SSI's launch, he said “superintelligence is within reach,” adding that “we are pursuing safety and capability in parallel.”

Sutskever has the credentials to aim for such advanced models. He was a founding member of OpenAI and previously served as the company's chief scientist. Previously, he worked with Geoffrey Hinton and Alex Krizhevsky at the University of Toronto to develop “AlexNet,” an image classification model that transformed deep learning in 2012. This development, more than any other, helped spark the rapid growth of AI over the past decade, demonstrating the value of parallel instruction processing by graphics processing units (GPUs) to speed up the performance of deep learning algorithms.

Sutskever is not the only believer in superintelligence: SoftBank CEO Masayoshi Son said over the weekend that “in 10 years' time, we will see the emergence of AI that is 10,000 times smarter than humans,” adding that making ASI a reality is now his life's mission.

AGI within 5 years?

Superintelligence goes far beyond artificial general intelligence (AGI), which is also a hypothetical AI technology. AGI will exceed human capabilities in most economically valuable tasks. Hinton believes AGI could arrive within five years. Ray Kurzweil, Google's chief scientist and AI pioneer, defines AGI as “AI that can perform any cognitive task that an educated human can perform.” This will happen by 2029. But in reality there is no universally accepted definition of AGI, and it is impossible to accurately predict its arrival. How will we know?

The same could probably be said about superintelligence, although at least one prophet has said that superintelligence could arrive shortly after AGI, perhaps as soon as 2030.

Despite these experts' opinions, it remains an open question whether AGI or superintelligence will be possible within the next five years, or even if it will ever come to fruition. Some, such as AI researcher Gary Marcus, believe that AGI (and even superintelligence) will never be possible with the current focus on deep learning and language models, and that these technologies are fundamentally flawed and weak, allowing progress only through brute force of data and computational power.

Pedro Domingos, Professor of Computer Science, University of Washington, author Master Algorithmis superintelligence A dream“Ilya Sutskever's new company is sure to succeed because superintelligence, which will never come to fruition, is guaranteed to be safe,” he wrote on X (formerly Twitter).

What's next?

One of these views may turn out to be correct. No one knows when AGI or superintelligence will emerge. As this debate continues, it is important to recognize the gap between these concepts and current AI capabilities.

Rather than speculating only about far-off possibilities that fuel stock market fever dreams and public anxiety, it is at least as important to consider the more imminent advances that will likely shape the AI ​​landscape in the coming years. These developments may not be as sensational as the grandest AI dreams, but they will have a profound impact on the real world and pave the way for further advances.

Looking ahead, we expect to see AI language, speech, image, and video models (all forms of deep learning) continue to evolve and proliferate over the next few years. While these advances may not lead to AGI or superintelligence, they will undoubtedly make AI more capable, useful, reliable, and applicable.

That said, these models still face some significant challenges. One major drawback is their tendency to occasionally hallucinate and confabulate, essentially making up answers. This unreliability is currently a clear barrier to widespread adoption. One approach to improving AI accuracy is retrieval augmentation generation (RAG), which integrates current information from external sources to provide more accurate responses. Another approach is “semantic entropy,” which uses one large language model to check the behavior of another.

There are no universal answers about AI yet

As bots become more trustworthy over the next year or two, we will see more and more of them being integrated into business applications and workflows. To date, many of these efforts have not lived up to expectations. This is not surprising, as incorporating AI represents a paradigm shift. In my view, it is still early days and people are still gathering information and learning how to most effectively deploy AI.

Wharton professor Ethan Mollick agrees. Useful things Newsletter: “Right now, no one, from consultants to general software vendors, has the universal answer for how to use AI to unlock new opportunities in a given industry.”

Mollick argues that much of the progress in implementing generative AI will come from employees and managers applying the tools to their areas of expertise to learn what works and adds value. As AI tools become more capable, more people will be able to improve their work outcomes, creating a flywheel of AI-powered innovation within the enterprise.

Recent advances demonstrate the potential of this innovation: Nvidia's inference microservices can accelerate the adoption of AI applications, and Anthropic's new Claude Sonnet 3.5 chatbot is said to outperform all of its competitors. AI technology is finding applications in a wide range of fields, from classrooms to car dealerships to the discovery of new materials.

Progress will steadily accelerate

A clear sign of this acceleration is seen in Apple's recent announcement of Apple Intelligence. As a company, Apple has a history of waiting to enter the market until a technology is mature enough and there is demand for it. This news suggests that AI has reached an inflection point.

Apple Intelligence goes beyond other AI announcements by promising tighter integration between apps while maintaining user context to create highly personalized experiences. In the future, Apple plans to allow users to implicitly combine multiple commands into a single request. These commands may be executed across multiple apps, but will appear as a single result. This is also known as an “Agent.”

During the Apple Intelligence launch event, SVP of Software Engineering Craig Federighi walked through a scenario that shows how these features work: As reported by Technology Review, “Suppose someone emails him saying a work meeting has been postponed, but his daughter is in a play that night. His phone will now find a PDF with information about the performance, predict local traffic conditions, and let him know if he'll make it on time.”

This vision of AI agents performing complex, multi-step tasks isn't unique to Apple — in fact, it represents a broader shift in the AI ​​industry toward what some are calling the “agent era.”

AI is becoming a true personal assistant

In recent months, there has been growing industry discussion about moving beyond chatbots into the realm of “autonomous agents” that can perform multiple related tasks based on a single prompt. These new systems would use LLMs not just to answer questions or share information, but to complete multi-step actions, from developing software to booking flights. According to reports, Microsoft, OpenAI, and Google DeepMind are all prepping AI agents designed to automate more difficult, multi-step tasks.

OpenAI CEO Sam Altman described his vision of Agent as “an incredibly capable colleague who knows absolutely everything about my life, every email I've ever had, every conversation I've had, but who doesn't feel like an extension of me” — in other words, a true personal assistant.

Agents are also useful for enterprise-wide applications. McKinsey Senior Partner Lari Hämäläinen describes this advancement as “software entities that can orchestrate complex workflows, coordinate activities across multiple agents, and apply logic to evaluate answers.” These agents can help automate processes within an organization or augment employees or customers who carry out those processes.

Startups focused on enterprise agents are also emerging. Emergence, for example, has just emerged from stealth mode. According to TechCrunch, the company claims to be building an agent-based system that can perform many of the tasks typically handled by knowledge workers.

The way forward

The emergence of AI agents will enable us to participate more effectively in our always-connected world, both for personal and professional use, as we increasingly interact and interact with digital intelligence everywhere we go.

The path to AGI and superintelligence remains shrouded in uncertainty, with experts divided on its feasibility and timeline. However, the rapid evolution of AI technology is undeniable, and transformative advances are expected. As businesses and individuals navigate this rapidly changing landscape, the potential for AI-driven innovation and improvement remains great. As the line between human and artificial intelligence becomes increasingly blurred, the path ahead is as exciting as it is unpredictable.

By planning proactive steps now towards investing in and engaging with AI, upskilling employees, and ethical considerations, businesses and individuals can position themselves to thrive in an AI-driven future.

Gary Grossman is executive vice president of Edelman's Technology Practice and global leader of the Edelman AI Center of Excellence.