AI and the Innovation Impasse
AI has taken menial work off my plate but some of the joy of development with it. When it comes to new ideas, AI just defaults to generic results which are seen everywhere else.
Image source, Magnific When I’m not sat writing articles (yes, these are not AI written posts), my day-to-day job as a developer has been one of great evolution. I could picture myself as the modern-day Odysseus in a new tale of the web developer navigating AI. Perhaps I’ll put it to Paramount to see when we can get the ball rolling.
Ever since AI was introduced in my workplace, I began a love/hate relationship with it and embarked on a long journey with it.
On one hand, I am able to deploy AI to do the menial, soul-sapping work I managed for years, the kind of work that should have been automated when IFTTT was first launched. In doing so, I have freed up my time to focus on more pressing matters.
On the other, AI has kyboshed my joy of developing. Where I would once find a sense of accomplishment in solving something, now I find myself moving from task to task babysitting an AI.
Many a moment, like now, I find myself reflecting on what development was like. While I do code, there’s this strange pull, not unlike that (I imagine) experienced by a gambler, to give the machine one more spin. Yes, I can build, but so can the AI (assuming I prompt it sufficiently), faster. I will, however, state that for now I hold the upper hand in error-free delivery.
On a personal note, it has been a bittersweet situation with AI. Perhaps it’s the environment I’ve found myself in. While I don’t regard myself as requiring constant affirmation, when you work hard to deliver something which you put your heart and soul into, it is regrettable to not have any feedback.
In situations where that effort and countless hours stressing is effectively overlooked, you start to think of letting the AI do the legwork. Yes, you can verify it, but otherwise, to coin Star Wars, “let the wookie win”. Why work harder when you can offload the work and focus towards writing smarter rules for the AI to follow?
Thing is, occasionally you have to put the effort in. In fact, no situation is less vital than when trying to innovate. Here is where I find my influence remains head and shoulders above AI.
Let’s play out a scenario. You work for a newspaper in their development team. You have been asked to think of new features you can implement. You ask AI and it hands you a bunch of suggestions. Some of them are probably great and fit right in: bespoke newsfeeds, smarter algorithm article matching, custom preferences and the like.
Ask yourself and be honest: do the above suggestions strike you as innovative? No. They’re common among similar providers. AI is trained on data, so its suggestions are also based on that data. As such, rarely will anything new be spat out as the answer. You’ll not see Metaverse theatrics or a 4D sensory experience. Imagine smelling bread from a bakery website… I’m going out on a limb and imagine it is less likely any of these would show in a basic prompt response.
There will be a few of you highlighting that 4D isn’t new, feasibility would be unlikely, la-di-dah. The point I’m making is that ‘ideas’ remain surprisingly human.
So, where does that leave AI and innovation? Remarkably, the answer is ‘at an impasse’. AI cannot reliably provide unique enough answers; instead, it uses the data it has aggregated and finds a ‘best fit’.
Image source, Magnific If you ask an AI for a random number between 1 and 100, it has a greater chance of responding ‘73 or 42’. If you have ever watched the TV show ‘The Big Bang Theory’, Sheldon (one of the main characters) declares the number 73 as the best number. Seems there’s a bias towards that number as it’s a popular pop culture reference.
Going by that logic, I greatly anticipate the data AI is founded on would have a significant amount of bias. As such, innovation from AI could be constantly handicapped.
Perhaps a few sessions exploring this might change my mind.
Has AI ever handed you an idea you’d genuinely call new? Or does it keep returning the same well-trodden feature list dressed up differently? If you’ve found a way to prompt your way past the ‘best fit’ problem, I’d like to hear it.
