Avoiding a Colossal Wreck

Humans In The Loop

"Round the decay
Of that colossal Wreck, boundless and bare
The lone and level sands stretch far away.” Percy Shelly - 1817

With apologies to Percy Shelley, we thought his poem had some application to the generative AI landscape today. We’ve seen the passions of many app creators declaring the arrival of a whole new era of computing with generative AI; however, we remain skeptical about real productive outcomes for business given the current state. Of course, the GPT models from OpenAI, Google, Meta and others are astounding, delivering creative prose and completing search at staggering speeds. But on closer inspection, it's apparent that most AI business apps are relegated to repetitive, low-stake tasks. The impact remains muted on productive ‘needle moving’ business outcomes for one colossal reason: misinformation.

We suspect you’ve read the headlines about calls to slow down AI research in the USA and increase government regulation over the industry (and you can imagine how that will turn out). But amidst all the clamor, our development lab has demonstrated the challenge of directing generative apps to stay on target and address intended functions and outcomes. We’ve worked to throttle output, audit results, align generation and control quality through programmatic means. Alas, the risk of misinformation is high, and a breach of trust in any production process would leave a colossal wreck. You can read more about our observations here and here. Even Geoffrey Hinton, the renowned AI pioneer, understands this risk, and has left Google to focus his energies full-time on the issues.

So where do we go from here? If you have been tracking our posts, you know that we believe in the promise of this technology, but only with thoughtful deployment of effective methods to direct outcomes. The most effective method we have seen to-date is RLHF, which is an acronym for ‘reinforcement learning from human feedback”. Another way we like to think of this is ‘human in the loop’ processes, where AI takes a secondary role providing intelligent workflow.

We’ll have much more to share on this topic as we gain additional insights from client engagements. At Strategic Machines, we know every company is looking for a colossal victory.