The problem of AI alignment is one of the most important questions we need to answer to safeguard humanity’s future. How do you ensure that an Artificial General Intelligence will behave ethically?
I outline a general approach to achieve this goal that counterintuitively relies on confusing the AI on purpose.
This approach relies on a number of basic observations about the nature of Artificial Intelligence.
An AI is different from a human in multiple ways. This is part of what makes AI alignment such a difficult problem, because our intuitions for how people act often do not apply to AI’s…
(This article tries to make concepts from machine learning digestible to non-technical readers. I try to keep things non-technical.)
I am continually amazed by how often I find analogies between my work as an AI researcher, and my own thoughts. When working on an AI algorithm, nothing is a more promising sign than the realization that my own brain has the same behavior as the algorithm I am developing.
This is especially true when the behavior is irrational. …
I invented a new type of layer for neural networks, and I’m hoping for feedback on it.
The new layer gives the network the ability to assess the reliability of its own features. This allows it to change how much weight it puts on each feature depending on the problem.
When a normal neural network solves a regression problem, it only provides the raw output. It does not provide a measure of how certain it is of the correctness of that output.
This is a problem if the network has no idea what the output should actually be. It will…
Even well-designed AI systems can still end up with a bias.
This bias can cause the AI to exhibit racism, sexism, or other types of discrimination. Entirely by accident.
This is usually considered a political problem, and ignored by scientists. The result is that only non-technical people write about the topic.
These people often propose policy recommendation to increase diversity among AI researchers.
The irony is staggering: A black AI researcher is not going to build an AI any different from a white AI researcher. That makes these policy recommendations racist themselves. …
This article explores the so-called “Hardest Logic Puzzle Ever”, and shows how thinking outside the box can sometimes take you a lot further than you would expect.
The Hardest Logic Puzzle Ever is a logic puzzle so called by American philosopher and logician George Boolos and published in The Harvard Review of Philosophy in 1996.
It is stated as follows:
I have invented a new financial instrument based on crowdfunding that is specifically designed for fast-growing tech startups.
Feedback has been overwhelmingly positive so far.
Unfortunately, actually implementing this system would be a lot of work for me and I don’t have the time to spare. I’m hoping that someone else will implement it, and I can join them as their first customer, with the tech startup I am building (elody.com).
This financial instrument has a very different risk/reward system than traditional crowdfunding methods: Instead of a large chance to lose all your money and a small chance to make…
In my work as a data scientist, I have noticed that many tasks that used to be difficult keep getting easier because of automation.
For example, AutoML promises to automate the entire model-building process.
While that is amazing, the work of a data scientist is much more than just implementing a machine learning model.
As it turns out, the aspects of data science that sound the sexiest will be the first to be automated, and the ones that are the hardest to automate are the ones you would least expect.
What most people focus on when you talk about data…