Thread by @BMarcusMcCann

I teach machines to learn from mistakes.

I don’t see fear or anxiety in that process.

I do see fear and anxiety in how a lot of people relate to mistakes.

Should we train ourselves more like AI?

Should we train AI more like ourselves?

🧵

First question.

The role that mistakes play in learning has been a key theme of how I’ve been trying to understand language, learning, and conceptual shifts (both in individuals and in groups of individuals).

People are often more creative when they are allowed the space to freely make and learn from mistakes.

Even in high-pressure situations when necessity provides the impetus, creating a space in which mistakes are framed as learning opportunities leads to better outcomes.

These spaces provide more opportunities to iterate, refine, build intuition, and grow.

As people grow, they are able to take on more difficult challenges while maintaining these spaces. They recognize mistakes but also appreciate their benefits.

As I grew as an AI researcher, I became more comfortable maintaining a play mindset with respect to goals that required more time, more compute, more collaborators, and more debate. As a CTO, I see it as my responsibility to now enable my colleagues to enter such a play mindset.

These lessons have, for me, also transferred quite well to relationships with family and friends. By lightening the gravity of mistakes, we work better together, especially in situations that are tense or have important financial or psychological implications.

We accept that we’ll still make mistakes, but we focus on iterating together towards a better outcome — just like in machine learning where errors exist, but they don’t discourage iteration.

This takes time and supportive, trusted relationships — the kind in which you can make mistakes and you both know it’s going to be OK. You can laugh about it because, in the grand scheme of things, you’re playing together, creating together, and growing together.

If people don’t have those supportive relationships, they can get weighed down by fear of mistakes, which can stifle their creativity.

We can take a lesson from machine learning — treat mistakes as learning opportunities and create spaces for free, playful thinking.

Second question. Should we train AI more like ourselves?

While it’s good to have that space to play and think creatively, there is certainly a positive role those feelings of fear, anxiety, and discomfort play as well.

Humans experience discomfort when they aren’t fully prepared to carry the weight of potential mistaken decisions. That gravity we feel is a signal that we’re either stretching ourselves, completely out of or depth, or more likely somewhere in between.

Stretching → like a good workout providing stimulus for growth.

Out of your depth → consult others.

I use those feelings as indicators for how prepared I am and how much help I need. Then, I use that information to create an appropriate space of play for myself and collaborators, where for some amount of time, we throw off gravity and freely create the best solution.

Unfortunately, I haven’t yet seen a good way to get AI to be more human in this respect.

AI has no sense of which decisions carry greater risks or deserve the additional gravity — the humans always decide for them one way or another.

It can be risky when humans offload too much of that gravity, too much of the weight of decision-making onto the shoulder of AI.

There are decisions that deserve the reverence and respect that only humans can feel.

It is important that we do feel it and carry that weight ourselves, in part by holding it up high above our heads so that there is plenty of space for free, creative thinking.

I write about the intersection of philosophy + technology. If you’d like to learn more, follow me @bmarcusmccann.

You can also see my work in progress as I build you.com @YouSearchEngine, a new private search engine.