An extract from our upcoming book, Leading the Open Enterprise:
We can’t have a section on technology without discussing AI. Like every new technology before it, it has been hyped to absurd levels by the vendors and the commentators. AI is a particularly egregious example because the business model seems to be to burn large amounts of venture capital, pump the share price, and run. Nevertheless, it is real. At Teal Unicorn, LLMs have been transformative to our work. We use them daily, mostly ChatGPT but also Dall-E and Gemini.
For years we battled to stop people saying “AI”. There’s no intelligence here. It’s an astonishing programmer party trick, an amazing simulation. But we gave up, and now we use the term. More correctly, we should say “machine learning”, and more specifically “Large Learning Models (LLM)”, or even more specifically “Generative Pre-trained Transformer (GPT)”. But hey, it’s all AI. And it blows us away. We have gone from sceptics to using AI every day. We use AIs to
- get us started with an abstract or a list of bullet points on some topic
- translate content to or from Vietnamese
- analyse content and summarise it
- write content for us
- edit our content for grammar, structure, logic, or brevity
- create graphics
- design articles, presentations, courses (this one!), or books
- consolidate content into an interactive body of knowledge
We regard AI as a form of work automation. There are some principles of automation that we can only go briefly into here, which guide us in our approach to AI, for ourselves at Teal Unicorn, and when advising clients.
All automation makes you superhuman: you can achieve more with it than without it. Often you can do things that without it you can’t do at all. A farmer can harvest a field alone with machines. A factory worker can shape metal in seconds. A software programmer can create binary code using high level languages then test and deploy it in seconds. Bank customers can do their own banking online. A fighter pilot can fly a plane that should be unflyable. A commercial pilot can land in a fog. A researcher can instantly read documents stored in a distant country. Automation is like wearing an exoskeleton: it enhances how you work.
Automation is expensive, fragile, and constrains adaptability. Automation is not always the best solution.. You’ve got to expend a lot of money and effort to bound the system so tightly that it can be simplified until it can be automated. If the conditions change, automation can’t change itself (not even AI automation), especially if the system goes chaotic. If all hell breaks loose, humans have to sort it out.
Automation doesn’t stop all errors. They are just less frequent and bigger. Automated systems are tightly constrained with as much error prevention built in as possible. But nothing is perfect. When the automation breaks, the results are usually catastrophic. It is expensive to clean up the mess and restore operations. Look up videos of roller steel mill failures. “To err is human. To really screw things up you need a computer”. In some cases manual fallback is impossible: you’re out of business until you get it running again. The good news is that automation is antifragile: you can add more automation to stop that happening again.
All automation depends on the user using it responsibly. One human error annihilated the stock trader Knight Capital in 20 minutes. Cars don’t have a big sign on the dashboard saying this vehicle may kill you. LLMs don’t present what they say as truth. They’re covered in disclaimers. Yet lawyers and politicians keep publicly humiliating themselves by uncritically using the output as fact. Every tool has hidden costs that people learn through bitter experience, and eventually becomes social lore. The public are rapidly learning not to believe AI without first applying some critical thinking. Just like google search or asking a colleague.
Automation changes everything and nothing. AI will be as transformative as the internet was, which was in some ways a lot and in others not much. Operating models and products change quickly: human aspects of business, less so, and more slowly.
Your staff aren’t going away. It is very rare that automation eliminates workers entirely: there may be less of them, and they may work in more skilled higher-value work, but they don’t go away. Every attempt at “human-less factories” has failed: they all revert to having some staff. A few professions have disappeared entirely, like wheat harvester or telephone operator, but unemployment has varied around 5% for centuries, and companies haven’t changed size either. What AI isn’t going to do – and shouldn’t – is radically reduce or replace your staff, so don’t start there. Remember, staff are an asset not an overhead. People worry that there will be social disruption from AI putting people out of work, like coalminers in Wales or auto workers in Detroit [but it wasn’t automation that did it to them, it was social change]. There potentially is, and we are not trying to trivialize it. But we don’t think the impact of AI will be anything like as painful. This time it’s not manual labour, it’s knowledge work. Service workers tend to be able to reskill and reinvent better. They will exploit the automation in their jobs, not lose to it. You won’t lose your staff – and don’t you even think of replacing them. Just like all automation, AI will make them more efficient and effective, freeing them up for higher value work. We’ve never seen knowledge workers who didn’t have a huge backlog of higher value work which was not getting done. AI will make your staff better, not fewer.