The hype, the danger, and the real promise of AI

We are now at peak AI hype, and I would like to just give you 10 quick things you did not know about how to use ChatGPT… Just kidding, I would like to have an informed real debate about the moment we are facing in technology right now.

I’ve spent the last 15 years building large-scale supply chain software for the enterprise at Tradeshift, after working with the Danish and EU governments, and from a background studying language and emergent complex systems in Sociology, so I feel I have a good position to understand the unique technological, ethical, geopolitical and business concerns and opportunities this moment poses.

This has given me a unique perspective, and if we’re to get real about AI, and specifically Transformer-based Large Language Models (LLMs), I want to talk a little about what opportunities we are facing today.

—

The first thing to say is: your chatbot is not going to turn into Skynet and take over the world. As ever, humans remain the biggest threat to humanity. I believe that AI technologies hold the most promise to solve real threats to our planet like global warming, pandemics, or food shortages.

The reality is those who currently want regulation either don’t understand the technology well enough to know where to start or understand it exactly so well that they know regulation is probably the only moat that will protect them in a world where LLM technology will diffuse everywhere (the original transformer model is 200 lines of code and I got an open-source version running on my laptop in 10 minutes that is close to being as powerful as ChatGPT 3.5).

Meanwhile, the positive hype is also a problem because it’s making people skeptical and/or afraid of how these technologies might actually change their day-to-day lives.

The most important thing to understand is how they bring down the cost of innovation. Instead of booting up the Xbox, you can, as I did this month, build a game with your child in a week.

What will happen when fields like programming and design are put in the hands of more people than ever before? What will be the impact on society as our cost of innovation drops closer to zero — good and bad?

An asteroid scrolling game I made with my son in one weekend, using ChatGPT and Midjourney in a programming language I never used before, shows how rapidly the cost of innovation is going toward zero.

—

In the mid-20th century, the shipping container transformed the world by giving every manufacturer access to the global supply chain. This created whole new economies of scale, gave developing markets an “on-ramp” to global trade, and raised standards of living for billions.

These LLMs can be the container for creativity. They unlock the possibility for anyone with a basic computer and a little ingenuity to compete in the global marketplace for ideas — with near-perfect translation.

For now, the fact remains: the potential of these tools is still defined by human supervision and collaboration. Simple second-generation tools like AutoGPT, which automates combining tasks through ChatGPT, look great in a demo — but in the real world, even small errors quickly cascade into a repeating loop that makes results unusable. What may impress in a Twitter demo can be a completely different experience in the wild.

And yes, (now as ever) this also means bad people can use this same new technology for nefarious purposes. I understand the threat, and we do need to find solutions — by strengthening our public institutions, education, and civil society. A ban will only affect the legal and desirable users, while bad actors are driven underground.

—

For me, this opportunity around human creativity is just the start. In the near future, the real promise is to solve a problem that plagues billions of humans every day:

Modern work sucks.

In the discussion of a looming “white collar job apocalypse”, too few are asking: how much of this is really the work we want to be doing if we could choose? Is much of it even human?

The darker reality is that for many, maybe most, there is too much of the work we have to do, versus the work we want to do. Repetitive, unengaging tasks and admin, usually assigned to you by someone else.

Most work software is not made for the workers, it is made for the managers. It makes life easier for the HR Manager if you log in every day and do approvals. And older software is even worse — does anyone get joy out of the existence of SAP or Salesforce other than shareholders tallying the profits?

In most cases, people are not using the software, it is using them to further the purposes of others. Click a button here, change a bit of data there… and if you’re lucky it will trigger something for someone you don’t know or see — and then, maybe, just maybe, you can get on with your real work.

This is also why the answer won’t be adding yet more to the existing interface — (Application “copilots” in enterprise software will soon feel like Microsoft Clippy.)

This goes against the hope most of us had coming into technology, that we could use this magical thing to make the world better, to create opportunities. There is a reason that the hippie movement and the first Silicon Valley dreamers had such a big overlap, they all dreamt of something better. Was Steve Jobs really the last dreamer and what would he think of our enterprise applications today?

So where did we lose our way?

—

Language is the most powerful tool we have ever created for understanding and communicating with each other.

What if we could redesign computing so that language is also the most powerful way to communicate with computers? What if the future of technology mirrored the social skills we have practiced for millennia?

This is where LLMs offer true promise. They can replace over-designed, sprawling user interfaces with something much simpler and more human.

Just tell your computer what you want it to do. Not with a keyboard and mouse, but in the way you interact with everything else in your life.

This next-gen of computing experiences will be both simple and powerful. Take Photoshop vs Midjourney.

One has gigabytes of features, settings, buttons, and toggles — and is still limited by your practice and experience. The other is a chat interface in Discord that lets the average, inexperienced user achieve results beyond what they had even anticipated. Instead of putting a lot of time and energy in, only to get little out, the equation will be reversed.

This will enable more creativity, more joy, and allow us to focus on the work we want to do, not the work we have to do.

—

Yes, this will change employment, just as employment has changed again and again throughout human history. We must be real and stop lying about it in a way that will only cause more distrust. Should we really keep a bad workaround that people have no choice but to engage with? Didn’t we move on from workhouses and other archaic ways for the same reasons?

What if the answer is to embrace a world in which those who are less willing or able to work have a choice? There are demographic shifts challenging countries like Denmark and Japan — we can offer more flexible retirement or a better work experience for those who want to work.

There are those who never wanted to do some of the jobs they are being forced to take today. As a society we need to decide on a path beyond work as we know it — not bury our head in the sand.

And meanwhile, it is very likely the satisfaction and reward of the technologies will help others achieve much much more.

The history of the next decade will be written by the choices we make now. The day-to-day lives of billions can be re-defined and improved.

Not focusing just on efficiency — but on humanity.