Beyond AI

Euphoria for new technology will ebb and flow. In the early days of AI, it was widely believed that we could build artificial general intelligence (AGI) which mimics the biology of the human mind and its synapses. In this form, the artificial “brain” could grow and adapt based on its own experience and outcomes.

In those days, the computing power was not adequate to come close to this vision, and we passed into what was considered an “AI winter” when funding (especially government funding) was significantly cut in favor of more achievable pursuits.

Later innovations in AI followed a path closer to the mathematical structures of the logic and circuits upon which modern AI was eventually built. These generated optimizations through techniques like gradient descent (to reduce error in supervised learning) and distances as a measure of similarity (to define clusters in unsupervised learning). Upon these foundations we’ve since built recursive and ensemble approaches with great complexity such as neural networks that are now coming close to achieving “human-like” outcomes.

At this moment, when AI can effectively drive our cars in many if not most circumstances, and (through new technologies like ChatGPT) seem to think, feel and empathize, AI is often passing the Turing Test–passing as a human in an interaction with a human. It’s forgivable then for one to think that we might be approaching a moment when AI can surpass human intelligence, an oft told trope in the daily media deluge on the topic of AI.

In an October address, acclaimed Stanford professor of computer science and Co-Director of the Human-Centered AI Institute Fei-Fei Li summarized this concept succinctly: “AI should be augmenting and not replacing human capabilities.” Looking at AI through this lens, AI becomes, even in fully automated systems, the recommender, giving its human trainers and operators the final say, all the while endowing humanity with the ability to increase both the velocity and quality of our endeavors.

In this regard, augmented reality (AR) and virtual reality (VR) may have been false starts. While there are certainly practical applications for these technologies, they do not represent the end state for Human+AI systems, rather just one use case, by adding information to the human field of view (AR) or offering an alternate viewing space (VR). With AR, we can see buildings that don’t yet exist or additional information about the built world in the context of reality. With VR, we can defy the limitations of physics and reality to be nearer to those we work with and access worlds limited only by our imagination. However, these scenarios are primarily focused on consumption, whereas most enterprise applications are in the domain of creation. For enterprises operating in the real world of creating and delivering products, services and experiences, Human+AI systems that truly deliver, are those that snap seamlessly into existing processes like demand forecasting, resource management, manufacturing, shipping, maintenance, customer support, and more.

At the advent of the internet in the mid-nineties, companies were grappling with how to integrate the web into their existing businesses. Analog incumbents in retail scrambled quickly to build an online presence to combat the increasing market share of online-only retailers. E-commerce was seen as an additional revenue stream vs. the traditional brick- and-mortar sources. Today, we rarely differentiate e-commerce from commerce. It’s just commerce.

Similarly, cell phones were initially an additional way to communicate (albeit more expensive) than traditional landlines. Now, we don’t differentiate between cell phones, they’re just phones.