When Will AI Be Good Enough?

And how will AI being good enough affect tech transfer?

Recent hype around ChatGPT and other LLMs (large language models) has resulted in two types of stories. The first type insists that generative AI will replace everything and everyone. The second type insists, just as vehemently, that these new AI models will never be good enough to replace or even augment any human activity – that the entire hype cycle is just a mirage. This second camp would chuck all AI advances in same bin as blockchain and other technologies that have failed to be widely adopted.

Neither side seems interested in the question of which human activities these new AI models could be good enough to successfully replace. Focusing on replacement of specific human activities with AI is much easier to imagine. As I mentioned in my last newsletter, even IBM is only considering replacing certain back office functions with AI. The only reason that would lead to massive job losses is because IBM is such a large company that many workers are focused on very narrowly specific tasks.

When is enough really enough for AI?

The next question becomes, How will we know when an AI enabled technology will be good enough to replace a specific human activity? What is “good enough”?

David Karpf, a professor at George Washington University, recently brought up the concept of “satisficing” in the context of AI in his Substack newsletter (link below). He states that while generative AI can provide a workable solution, for example to plan a vacation, it can’t provide the best solution. “Satisficing” is the idea that we all seek such “good enough” solutions in our daily lives. Prof. Karpf goes on to argue that these “good enough” AI based solutions will have limited applications and will ultimately prove disappointing for corporations.

But let’s consider regular human work in the context of a large corporation. Most companies exist as a collection of repeatable - and repetitive – processes performed by humans and supported by machines. Generative AI ups the ante through flexibility. As I pointed out in my last newsletter, the new AI models are very good at coping with the messiness of human life. “Good enough” AI should be able to replace many repetitive business processes. These processes aren’t optimal or optimized. They aren’t creative. And they certainly aren’t innovative.

Turning back to the case of IBM, it plans to eliminate nearly 8000 jobs with AI that can do such simple and repetitive tasks as issuing proof of employment letters. IBM doesn’t need super charged, optimal AI that can do many different tasks with creativity. It needs focused, reliable AI that can do one task well but with some flexibility. ?In other words, IBM needs “good enough” AI.

How will this affect tech transfer?

Generative AI would seem to open up many new areas of high level innovation. But large corporations don’t necessarily need or want high level innovation. They need reliable innovation which can fit seamlessly into existing processes, without causing additional headaches for existing workers – and which can help the company to increase productivity per worker, reduce the number of required workers, or both.

Think about how computers originally integrated into the workplace for large corporations. At first, they replaced secretaries for typing letters and filing cabinets for document storage. By replacing filing cabinets, computers also replaced the jobs of filing clerks. With the advent of email, the computer reduced or eliminated the necessity for fax machines and postal services – and those accompanying jobs. At each step, computers replaced more and more job functions – or even complete job categories – without requiring the large corporation to rapidly change its business processes. Gradually, adoption of computers did cause these business processes to change, as humans adapted their work to take advantage of the new technology, but this was gradual and relatively non-disruptive.

For tech transfer, this means that innovation requiring disruptive changes to business processes is not likely to be successfully licensed to a large corporation. In some cases, the large corporation may decide that it will tolerate or even welcome disruptive change. But it is far more likely to look for innovations that can provide incremental benefit. Think AI to replace humans sending proof of employment letters, rather than replacing the entire HR department.

What is "good enough" AI for tech transfer??Comment below, or?click this link to talk with me ?directly about your own tech transfer challenges.

References:

David Karpf’s Substack on AI and satisficing: https://davekarpf.substack.com/p/on-generative-ai-and-satisficing

This lawyer argues that AI legal tech should be accepted if it’s good enough: https://abovethelaw.com/2022/08/when-is-ai-good-enough/

#innovation #AI #ChatGPT #techtransfer