The Future of AI: A Repeat of the Past
Reading about the history of the integrated circuit in The Chip by T. R. Reid, there’s an interesting section explaining how Intel quickly realized that building different integrated circuits for hundreds of products simply wasn’t practical. Instead, Intel recognized that a single general purpose microprocessor programmed for different applications would be infinitely more practical and profitable. History has proven this approach worked.
Today, general purpose AI has emerged as a replacement for specialized machine learning techniques. While it may not be as fully capable of rigorous statistical analysis yet, trends suggest it will reach those capabilities soon. Why does this matter? Because general purpose AI (generative AI) may follow the history of the general purpose integrated circuit (the microprocessor). If it does, what can we expect? Here are four quick predictions I expect to play out.
Personalization
Similar to how the integrated circuit enabled the personal computer revolution, soon AI systems will enable personal AI revolution. The best way for AI to augment our intelligence is when it knows each of us personally. We’re already seeing this in foundation models use of memory and chat history. Now, companies like Lenovo released Qira for on device AI that personalizes it’s interaction for each user. This wave of personalization will continue, with battles between on-premise/edge systems and cloud systems competing on who can offer the best mix of features.
Client Server architecture
The personal computer revolution led to the client/server architecture, where local computers (clients) did some work and server computers did some work, split by what work could be accomplished on each most effectively and efficiently. Organizations saw huge improvements in system capability through this architecture. In later years, the client-server architecture has been partially replaced with cloud computing. However, that can be seen as a weak client architecture through the use of a browser. Regardless of semantics, I expect AI evolution to show some similar developments where personalized AIs for employees (clients) interact with organizational AIs (servers), to accomplish organizational tasks.
Multi-core
Later generations of the integrated circuit saw multi-core technology emerge to leverage parallel processing. We’re already seeing top AI engineers using multiple instances of AI to do multiple tasks at the same time. The key problem with this is that it requires multiple instances. In the not too distant future, a single AI interface will spawn multiple AI tasks, each working in parallel in the background to accomplish tasks.
System on a Chip
Another development of the integrated circuit is the development of a system on a chip, which takes multiple pieces of computing technology, such as memory management, CPU, GPU, IO controller, etc., and integrates them into a single chip. These chips enable smaller implementations of computing architecture without the need of many different pieces of technology on the motherboard. The AI analogy will come in the form of a system in an AI, as in the entire organizational system. Eventually we will get to a point where all the disparate software apps are consolidated into a single AI that orchestrates the entire organizational information flows. While the AI may not be the fastest or most efficient at individual tasks, compared to other AIs, it will cut coordination and integration costs.
The Future of AI
Predicting the future can be difficult, but if we see patterns from the past, it can help guide predictions of future innovations. This is especially true, if the purpose of the innovations remain the same. The integrated circuit enabled Intel to dominate the computer microprocessor market and launch the personal computer revolution. Which AI company will dominate the AI revolution? That will depend on how well they understand the success of past innovations to solve our problems and architect their systems to continue solving those problems.
