Matthew Berman on AI Adoption, Hype and the Case for Caution

I want to respond to an article a friend sent me about the supposed death of AI. Ted Gioia, writing as The Honest Broker, argues that demand is shrinking and that the main flourishing uses are spam, scams and deception. He makes some worthwhile points, but I disagree with his overall conclusion.

Gioia begins with Microsoft’s investment in AI and the failure of Bing to make a large dent in Google’s search dominance. Microsoft has invested heavily in OpenAI and is integrating AI across its products. But I do not think it expected the addition of AI to Bing to overturn established search habits in a few months.

I am an early adopter, and I use ChatGPT for questions that ordinary search does not answer directly: recipes, travel and coding, for example. As a coding assistant it has become part of my workflow. Many people still default to Google out of habit, and it may take time to discover where a conversational tool is useful.

The interface is still awkward, and integration into products people already use could change how often they encounter AI. We are at an early stage of this particular wave: ChatGPT launched in November 2022, less than a year before this video. Its early adoption is not the whole history of artificial intelligence.

Gioia asks why people have not abandoned search if results have deteriorated. He points to Reddit and TikTok; I would add Stack Overflow. These sites contain user-generated material that can be assessed through votes, likes or other community signals, rather than just content-farm pages.

I expect platforms with valuable community data to use it to build or train AI products. The appeal of a conversational system is that I can ask for an answer directly instead of opening many pages and sorting through ads. That possibility does not remove the need to check what the model says.

Gioia is right that people do not want answers full of errors or outrageous claims. In my experience ChatGPT often handles everyday questions well, though it can hallucinate; my experience is not a guarantee of reliability. I expect better data and future models to reduce these problems, but that is a prediction.

The article treats falling ChatGPT website traffic as a bad sign and suggests visitors should double every few weeks if the technology delivers. I disagree with the expectation of sustained doubling: a product cannot keep doubling indefinitely. Some early visitors were trying it out because of the initial excitement, and a decline in visits is not the same measure as the number of users.

It took me time to find my own use cases, and now I use ChatGPT daily. Other people may take longer. Looking at the article’s technology adoption chart, I see some earlier technologies rising, dipping and then growing again; I think AI could follow a similar path as useful applications and integrations emerge.

Some newer products had very steep adoption curves because people already knew related devices and technologies. A tablet, for example, arrived in a world familiar with smartphones, cameras and the internet. I see conversational AI as a less familiar experience, so I would not expect every curve to look identical.

I disagree with the comparison to unwanted AOL startup disks. I expect AI to be integrated into many everyday products and to make tasks easier and more productive. The delivery mechanism matters: the basic ChatGPT interface may not be the long-term answer, and I speculated that Apple’s eventual approach might offer another model.

Gioia asks who wants AI-written books, AI-made music or AI journalism, and who would trust AI in a critical job. Those are fair questions. In 2023 I do not think generated books or music can compete with human creativity, but I do not expect an overnight, binary replacement of artists.

A more gradual path is artists using AI to help write songs or authors using it as an assistant. In my own work I research, make notes and write a first or second draft, then use ChatGPT to help refine it. It participates in the process rather than originating the whole piece for me.

The article also says AI is useful for scams and spam. New technologies can be abused, and those uses must be addressed, but they do not exhaust the legitimate applications. Gioia cites a poll reporting that 72% of Americans wanted development slowed and 8% wanted it accelerated; such figures depend on the poll and its wording.

People have understandable fears about the dangers of AI, including fears amplified by AI company leaders. Wanting development to proceed more slowly and safely is not necessarily the same as wanting no AI at all. Likewise, concessions to screenwriters in the 2023 Hollywood strike can protect workers while AI still becomes a useful writing aid.

Gioia also cites reports of models getting worse at math. I had discussed observed performance changes and speculated that safety guardrails could affect quality, though that explanation is not established by the examples alone. I hope new techniques will improve safety without sacrificing performance.

Copyright is another conflict. Gioia points to objections from Universal Music and lawsuits against AI companies over training data; I also note reports of music companies exploring AI tools. Those are distinct issues: a company can object to unauthorized uses while investigating tools and arrangements intended to protect artists.

I compare the clash between technology and law to Napster and the emergence of digital music services. Napster itself disappeared, while later streaming services developed under different business and licensing arrangements. The analogy is about how disputes and technologies evolve; MP3 existed before Napster, and this history does not settle AI copyright questions.

Gioia lists ransom scams, malware, cheap AI music and student cheating. These are real concerns, and we need ways to deal with misuse. He also argues that public suspicion, weak performance in some uses and the possibility of job losses all weigh against the technology; I do not think those points alone show that demand will disappear.

In my own work, coding assistance, writing support and answers to questions have been valuable. Legal and regulatory obstacles are real, as is the risk of deception. We should build safeguards and keep criticizing the technology rather than assume either that all of its uses are harmless or that it should simply be abandoned.

Gioia suggests putting the genie back in the bottle. I do not see how that would work or why we should stop trying to improve the technology. Skeptics have a role in challenging AI and its risks, but I remain optimistic that, developed thoughtfully, it can make the world better.