Got a free pass at the last moment to the AI Engineer World’s Fair in San Francisco, found a cheap flight and a cheaper Airbnb. So here I am. Snagged a spot on one of the couches in the Expo Hall… am waiting for the crowd to thin out before I head to the buffet line for dinner.

The fact that they made scholarships to the event available based on need convinced me it would be worthwhile to drop everything and go. But this is still very much an industry event. I’ve had the vendor conversations that I need to have. (And was gifted a pair of branded free smartwool socks!)

The nice thing about deciding to go on such short notice is that I’m not here to pitch. More just to observe and listen.

Does anyone really want or need a robot barista? Microsoft appears to think so.

What I am mostly observing is a bad case of tunnel vision. People outside Wall Street and the AI industry itself don’t really understand how much negativity and mistrust exists around these technologies. Activists tend to call for regulation — and while this is needed, it will also put these tools out of reach for independent developers and small startups like my own.

But it hasn’t happened yet.

I am tempted to just put my head down and use Claude Sonnet to build as much high-quality algorithmic generated code as I can, while the getting is good. In all likelihood, that is exactly what I will be doing for the rest of the summer. But I would also like to outline a few ways in which AI can be used to solve problems that are otherwise intractable.

I have identified six:

Accessibility, Community Moderation, Content Curation, Consensus Building, Education, and Employment

(A3C2E, if you need a really clunky acronym.)

• Accessibility. Voice recognition is almost to the point of being usable in place of keyboard and mouse, but not quite. As somebody who lives daily with constant physical pain from repetitive strain injury, the evolution of this technology to allow hands-free coding and web browsing on a range of devices would be night-and-day. It isn’t there yet. But it certainly could be.
• Community Moderation, Content Curation, and Consensus Building. We don’t need more generated artwork, writing, and music. We are drowning in high-quality information already. What we need are ways to make sense of the incredible richness of artistic output already available online, and connect creators with patrons and enthusiasts. Ditto for online communities of all types, and for any type of process that seeks to identify shared opinions and experiences — rather than upvoting the most extreme and polarizing opinions (something that social media algorithms appear to have encouraged, even before the advent of AI). Full disclosure: consensus building through textual analysis is what my recent work centers on.
• Employment and Education. We need AI that creates jobs, and not just for the most highly skilled echelon of engineers. LLM’s are at a really interesting phase in their evolution right now. The best of them are at a level where they can mimic the output of an actual human. That is to say, they can respond in real time to questions and unique inputs, and generating meaningful replies. But they make mistakes. Organizations have a choice: build oversight and review into their AI workflows, or relegate AI’s to low value interactions — users who can’t afford personalized support — and leave them to deal with the consequences. But it doesn’t have to be this way.

Could job descriptions emerge along the lines of “AI Supervisor?” Would it be possible to build marketplaces and add-ons that specialize in matching human editors and graphic artists to put the finishing touches on generated AI content that is “almost but not quite” good enough?Yes, and yes. This is a different set of priorities than the current shibboleth of “AGI / Nobel Prize Level Intelligence” but it could be applied in practice with genuine and immediate results.

As remote learning becomes more commonplace, the most valuable role of AI may not be as a search engine that provides results in complete sentences. Rather than an “encyclopedia-on-demand,” correctly trained and tuned models could function as a source of input, guidance, and feedback for students working independently, particularly adult learners and those re-skilling. Yes, in an ideal world most of us would like a human tutor or mentor to work with us, but for people who cannot afford to take time off or take out many thousands of dollars in loans to pursue job training, one-on-one AI training could become a valuable counterpart to video lectures and online multiple-choice quizzes.

The challenge is that none of the six categories above are easily monetizable. We are talking about meeting needs that are currently going unmet — rather than using AI to slash costs (in other words, eliminate jobs).

Given the timetable and realities of grant-based funding, charities and the public sector are extremely unlikely to lead this charge.

So what is to be done? I would advocate that it’s time for the industry itself to step up and invest in social impact projects.

This is not an exhaustive list. Other people may be able to dream up further use cases that are equally valid. But it will take more than a hack-a-thon or a single afternoon to generate meaningful results. As a coder who hails from a UX background, I am a firm advocate of user research. Don’t build vanity projects. Instead, find partners and stakeholders.

Without visible evidence that AI is contributing something positive to our world, public retribution will be swift and absolute. Regulation won’t end inequity or violations of privacy from AI, although it may curtail the worst abuses. It will lead to something more like a guild economy, and render another sector of software (like cryptocurrency and healthcare before them) off-limits to those without professional degrees and licensure, as well as a budget that includes full-time legal counsel on staff.

In the United States, we still have a window of time to build things that are meaningful and cool.

I don’t know how long it will last. Maybe six months? Maybe one year?

I took two days off to take the pulse of the AI industry, and see what I could learn firsthand. Now it’s time to get back to work.