Inside | Real news, curated by real humans
Inside AI

Inside AI (Sep 2nd, 2018)

Happy Sunday and welcome to the weekend edition of InsideAI.  I'm Rob May, CEO of Talla,  a new kind of knowledge base for Sales, Support, and Success teams that merges automation and machine learning with content.  I also host the AI at Work podcast - the most recent one is a fantastic talk with David Murgatroyd, who runs machine learning at Spotify.  

This weekend, we will look at the top articles and links from the week, then I am going to make a very unpopular and unusual prediction about AI.

Top links of this week:

A U.K. startup that uses AI to produce marketing copy plans to expand into the U.S. market. Phrasee is based in London and the company's clients include Domino’s, Virgin, and Superdry. Phrasee was founded in 2015 and has raised $4 million in a Series A round to support its deep learning and natural language generation technology. According to co-founder and CEO Parry Malm, their systems are able to write copy that is "indistinguishable" from copy written by humans, and can also predict whether the copy will be effective. The funding round was led by Albion Capital and will allow the company to launch in San Francisco in June. — THE TELEGRAPH

Singapore is launching two initiatives to train 12,000 people in AI. The programs are being spearheaded by AI Singapore, a national organization created in May 2017 to boost the country's AI development. One of the initiatives is called "AI for Everyone," and targets 10,000 secondary school students and working adults. AI for Everyone is free and supported by industry regulator Infocomm Media Authority (IMDA), software vendor Microsoft, and chipmaker Intel. The second initiative, "Foundation in AI," is geared toward 2,000 technically inclined professionals. Foundation in AI will focus on basic competency and software development skills for the Python platform. — ZDNET

Designers Philipp Schmitt and Steffen Weiss developed an AI system for creating a new style of chair. Called the chAIr project, the system has a generative adversarial network (GAN) trained on a dataset of 562 chair designs from the 20th century. The GAN network generates new images based on the dataset, and analyzes and refines the new images. The designers then used the AI images to make 3D sketches factoring in real-world variables such as the material used to produce prototype chairs. — FAST COMPANY

Google is open-sourcing its reinforcement learning framework. The company made the announcement yesterday, introducing the Tensorflow-based system that was designed for clarity and simplicity so that researchers can quickly prototype reinforcement learning algorithms. The framework is called "Dopamine" and is available today on Github. Google is also providing the  JSON data files so researchers can compare their data with that of agents trained in other networks, and a website so researchers can visualize the training runs. — VENTUREBEAT

  • Email gray
  • Permalink gray

-- Commentary --

This month I read a book called Factfulness, which points out how many of the trends that concern us in the world are things we are actually wrong about.  The trend from the book that most surprised me is the assertion that every major body predicting human population growth shows us leveling out at between 11 billion and 12 billion people in the world.  (Note:  I have no idea if this is true, and I don't understand exactly how it happens if humans live longer and longer but, the book authors are legitimate people so, I'm accepting it as true for the moment)

Assuming this is true, it's a good thing for the ecological sustainability of the planet, but it's a bad thing for an economic system like capitalism that is built on growth.  When the entire world population stabilizes, what happens?

Here is my prediction:   We start selling things to AIs.

In fact, if you think about where the world is going, AIs are going to do more work for us, and get more complex.  They are going to get individual identities.  They are going to proliferate aggressively, and as all this happens, these AIs will star to compete with each other.  I will create an AI and you will create an AI and there will not be a job for both of them.  How will the human (or AI) buyer choose?  The AIs will compete.

Once AIs are competing with each other for work - by reputation, by differentiation, etc, they will need ways to distinguish themselves.  In other words, they will have the same issues humans have - they want to appear better/different than others.  

So if you look out to 2050, the largest economy isn't the US or China - it's the bot economy.  Bots will go from having identity to having reputation to competing for work with other bots to wanting to buy things.  We will encourage it because it's a way to grow the population of individuals we sell things to.  

There is a really interesting book that is almost 15 years old called The Robot's Rebellion.  It walks the reader through the process of trying to  design system that will propagate genetic code.  It turns out, trying to design a machine that will survive in the world, given all the variability, is challenging.  So, as the world changes, this propagation system becomes more complex - eventually to the point that it becomes a learning/decision machine that does what is best to keep itself alive.  Organic algorithms went through this evolutionary process.  It's reasonable to think machine will as well.

So in short, here is the argument:

1.  As AIs get more complex, and proliferate rapidly because of the low marginal cost of making a copy, the designers of these AIs will need to differentiate them if they are to be successful.

2.  Eventually, designing the differentiation will become too complex, so, the AIs will be programmed to seek their own differentiation.

3.  This will lead AIs to make decisions to "buy" things like training, or other differentiators.

4.  This will coincide with a leveling off of the human population in the world, which is a bad thing for capitalism.

5.  Some people will have the bright idea to "sell" things to the AIs/bots.

6.  Steps 1-5 will advance until bots are the primary participants in the economy.

I hope that stimulates some interesting things to think about over your long weekend.  Happy Labor Day, and I'll see you next week.

@robmay

  • Email gray
  • Permalink gray

Subscribe to Inside AI