AI resources & ideas for the business-minded
In the last year, Artificial intelligence has moved from media hype and horror stories to having actual business results. Today, I thought I’d share the most important ideas and insights that I’ve collected in the last year on the topic.
In its basic definition, Intelligence means to choose (legere) between (inter) a pair or set of options
In the field of Artificial intelligence, technologists are focused on creating bots and algorithms that are as smart as or smarter than humans (regardless of the implications). Most AI systems work by optimizing with respect to some objective function, and we get to decide what the function should be. Think of Google Maps finding the fastest route to go from A to B.
a16z has released one of the best primers on AI http://a16z.com/2016/06/10/ai-deep-learning-machines/
NVIDIA made a blog post that explains the difference between AI, machine learning and deep learning (P.S. NVIDIA created the GPU that made deep learning possible with faster and cheaper parallel processing) https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-intelligence-machine-learning-deep-learning-ai/
Here are the 9 biggest misconceptions about AI https://medium.com/intuitionmachine/deconfusing-ai-and-deep-learning-20473d7578c0#.tuhj1o9us
Deepmind, which beat AlphaGo champions in 2015 and got acquired by Google, published a round up of their work in 2016 https://deepmind.com/blog/deepmind-round-up-2016/
Google TensorFlow and IBM Watson are ones of many tools that you can use to build intelligent software (TensorFlow is open source AI library https://www.tensorflow.org/, while Watson is a paid full AI+NLP platform https://www.ibm.com/watson/)
More and more tech companies are releasing their AI capabilities as platforms. Just look at Amazon https://www.wired.com/2016/05/amazons-giving-away-ai-behind-product-recommendations/
Speaking with bots is on its way to the mainstream
After the release of Siri, Alexa and Google Home, speech is becoming the better way of interacting with our devices, replacing keyboards and touch screens. It is more convenient to tell a bot who can access all our devices and apps what we want it to do so it execute tasks for us. Just take the example of Amazon Echo turning off the lights for us — a simple task that links the user who asks for this to happen, the Echo that processes this demand, and the light switch which is connected and ready to execute.
A post on HighExistence wrote “The day that I can reach out my hand and, with nothing but my thoughts, make my intentions manifest in the real world, will be the day that technology can be considered grown up.” We’re far from making the thoughts narrative come true, but speech is good enough.
Natural language processing (NLP) is the framework by which software can understand humans and chat with them. I wrote a post on some implications of NLP https://medium.com/@jadjamous/when-machines-can-understand-us-they-can-find-all-the-answers-7df19a4f22f1#.98n6kwxwb
The argument for widespread AI relies heavily on data and tracking
A common approach to train an AI system is to feed it massive data sets. Experts are however now working on enabling machine learning with smaller data sets. What we expect it to do in return is find patterns and make predictions.
On the importance of data:
Mobile is enabling a data revolution. The Internet of Things promises that every product or “thing” will have connected sensors that send and receive data.
But a lot of data will have to be manually uploaded
“Big data is all around you. Machine learning — whether that means chess-champion computers, autonomous cars, or face-recognition programs — is only as good as the data it studies (Recode). The new generation of artificial intelligences learn by training on data sets that are carefully curated by — wait for it — human beings. Today’s companies and organizations already have the information we need. But if machine learning is going to drive the next breakthroughs in a field like cancer research, it will have to unlock the wealth of data that’s currently tucked away in handwritten documents and paper reports. Paradoxically, the road to AI may lead through the dust of the file room.” — NewCo
And businesses will have to convince people to upload their data by providing them with personalized service in return, such as Spotify recommendations and Facebook newsfeed suggestions https://medium.freecodecamp.com/the-business-implications-of-machine-learning-11480b99184d#.v705dpbxw
We can increase innovation with more data,but some challenges exist https://medium.com/safegraph/where-should-machines-go-to-learn-c2461f7e45fc?mc_cid=1ab54eb1c3#.6vf4zv3ks
AI’s is first making its biggest impact in healthcare (as demonstrated by Watson Health and Deepmind Health). It will affect how we diagnose diseases and give medical treatments. Business Insider explains that an AI system recently diagnosed a rare disease that human doctors failed to diagnose by finding hidden patterns in 20 million cancer records.
It can also help in solving big public issues. By helping to allocate scarce public funds more accurately, machine learning could save governments significant sums. According to Stephen Goldsmith, a professor at Harvard and a former mayor of Indianapolis, it could also transform almost every sector of public policy.
“By 2025, AI systems could be involved in everything from population health management, to digital avatars capable of answering specific patient queries.” — Harpreet Singh Buttar, analyst at Frost & Sullivan.
MUST WATCH: Kevin Kelly on AI https://www.youtube.com/watch?v=Ttc-Jndmmz8&feature=youtu.be
“If AI can help humans become better chess players, it stands to reason that it can help us become better pilots, better doctors, better judges, better teachers.” — Kevin Kelly
MUST READ: Marc Andreeseen on AI http://www.vox.com/new-money/2016/10/5/13081058/marc-andreessen-ai-future
“We need to cross the AI Chasm to provide real value for customers”, says Simon Chang https://flipboard.com/@flipboard/flip.it%2FqMcJSi-crossing-the-ai-chasm/f-3684b42ebd%2Ftechcrunch.com
Finally, a great post on how to build an AI startup https://medium.com/startup-grind/building-an-ai-startup-realities-tactics-6e1d18a4f7ab#.edmgj4jof
My personal conclusions
I believe the next step in the evolution of the internet is moving from providing infinite streams of information to actually suggesting what information we need right now and what we need to do with each piece of information. The best way to get there is to be surrounded with devices that know people so well they can predict what everyone wants to achieve given their current mindset, their personality, their location etc.
There is real value in solving people problems in new ways. However, I think the most important question in AI is whether it will be a recommendation engine or a decision making engine — today it is still a recommendation engine most of the time. But wether AI will be making decisions for you without your input will be a major source of angst and buzz. We’ve already seen it with Facebook AI deciding what you should read and the described “echo chambers” that this entails. That’s scary, remember the AI in 2001 Space Odyssey that said: ‘I’m sorry, Dave. I’m afraid I can’t do that.’ The unknown threats to society are not to be taken lightly.
And of course there is the automation threat debate, which Stephen Hawking summarizes perfectly:
In a business system that favours low costs and maximum efficiency, machines are bound to replace humans. All we can hope for is that the fully-automated post-scarcity economy flourishes before the people who don’t own the robots become unemployed and unable to pay for all the amazing products that AI will help create.
Jad El Jamous
London, Feb 15th 2017