AI models give incorrect answers… as do humans.
Even across high education fields, humans give you wrong answers. Or more often, a wide range of half answers.
The AI is trained to imitate, with the goal to emulate. But that’s on error-prone humans. And yet many people hold AI to higher standards.
You can get better answers – from both AI and humans – by learning to ask better questions. This isn’t a new skill to learn, but with AI, you’ll need different context and framing. It’s an old concept with a new name… prompt engineering.
What’s the Difference with Creativity and AI Hallucinations?
They’re the same. Or more so, hallucinations are a subset of creativity.
Creativity: the ability to produce or use original and unusual ideas.
I’ve also argued for years that creativity is an illusion. It’s a truth that few people agree with.
Hallucinations: the ability to generate thoughts or output that aren’t based on reality.
When a human draws a unicorn, that’s both creative and a hallucination.
But of horse course, the domain changes the impact of a hallucination.
On paper, an AI model can generate a horse with a horn. And that doesn’t hurt anyone. However, an AI doctor could come up with a creative diagnosis and prescription. A prescription that could kill.
Data Limits for Training: Synthetic FTW
We’re running out of training data… I never bought into this argument. It misses a broader sense of humans and technology.
They’re correct that human output is limited. And we’re creating more at a faster pace, but it’s still a growing limit.
For non-linear thinkers, we see this as a red herring.
To understand the crux of my argument, let’s start with a base truth…
Anything we create is based 100% upon our past experiences, along with how we’re biologically built.
All creative work is a product of a system or many systems working together. That includes both mechanical and biological.
Many people don’t want to believe our actions are deterministic.
We’re complex organic systems… but systems nonetheless. And as we better understand any system, we can better predict outputs.
Updating humans takes longer than updating computing systems. And already, computers have exceeded human ability on many fronts.
Computers generate original content and ideas today. And we’ll continue to bake in more variance for a wider range of outputs. This divergence is a requisite for creativity. On top of that, we can give AI more dots to connect, along with better awareness.
Knowledge is compression. And this is core to information theory.
Synthetic data can already be more useful than human generated content. That’s one reason model distillation is becoming more common. For the neophytes, that’s using bigger models to train smaller models.
To move even faster we’re unshackling data collection from squishy humans…
Humanoid Robots for Human-Centric Sensing
Even with 8.2 billion humans roaming the Earth, we’ve only scratched the surface. We’re building better maps of our terrain, but have a long way to go.
To assist, we’re early stages with humanoid robots. They’ll help us better map and understand the world for a human point of view.
If a tree falls in the woods and no one’s around, does it make a sound. Of course it does, but that’s not the point of the question.
The sound isn’t sensed and recorded in a way that we can access. However, that all changes if we have robots wandering around for us… and alas, it’ll take a few decades before humanoid robots are commonplace in homes.
In the meantime, humans will still equip more sensors…
One new source of training data is already taking root. Meta has sold 2 million Ray-Bans, partnered with glasses giant EssilorLuxottica. These AI glasses come equipped with cameras to see as humans do.
Plenty of privacy concerns… but an unstoppable trend. I’m looking forward to audio reminders or floating text above people’s heads. That’ll help with remembering names.
In the U.S. alone, there are over one million people who are legally blind. And that number multiples if extending to visually impaired. AI smart glasses are going to unlock huge accessibility benefits.
As smart glasses continue to improve, we’ll gather incredible visual datasets to train on. But sight is just one of many senses to understand our world, for both humans and AI.
Sensing AI Beyond Humans
Human sight is also limited. For example, we only digest a small sliver of the electromagnetic spectrum. We can’t see radio waves, microwaves, and so on.
And I’m not talking about seeing these microwaves…
However, we’ve created different sensors and tools to “see” more of this spectrum. It’s been helpful to create better maps of our world. But AI is already better equipped to digest and use the data to find patterns and trends.
AI can connect dots we can’t see and will go far beyond our five senses.
Humans are largely in-the-loop today, but AI will push us further out-of-the-loop. And there’s the singular human brain to compare against AI, but you’ll level up by understanding human hive mind.
Want new ideas? Read old books.
Enjoy my winding thoughts on AI?
Did any ideas help you connect some new dots? Do you have criticism to share?
If so, I’d love to hear from you. Feel free comment or reach out anytime.