and Advance Artificial Intelligence
Artificial Intelligence, or AI, is the single most important technological trend in history. As I explained in my book The Singularity Is Near1 (Viking, 2005), thousands of years of gradually accelerating progress is leading toward a point in time when a computer can provide greater overall intelligence than a human (who is not herself enhanced by AI). I have consistently put that threshold at 2029.
In the 2030s we will merge our neocortex (the outer layer of the brain where we do our thinking) with AI in the Cloud (computation available through wireless communication) thus increasing our intelligence. According to my calculations, we will then multiply our intelligence a billion-fold by 2045, a prospect so transformative that we have borrowed a metaphor from physics and call it the Singularity. The metaphor is to compare this future historical event to the event horizon of a physics singularity (that is, a black hole) where it is difficult to see beyond the event horizon. The Singularity will profoundly transform every aspect of civilization—and we are among the lucky five percent of humans who ever lived who will have this happen within our lifetimes. It will, in turn, vastly increase our life spans.
Even before the Singularity, AI will have greater and greater impact on the way we live. Artificial intelligence follows my Law of Accelerating Returns, which observes that the price-performance, capacity, and impact of information-based technologies improve by a roughly constant percentage every year, thus expanding exponentially. This means that actual progress is slow at first, but builds up “momentum” over time, and finally progresses at explosive speed. We are just now reaching the point where progress in AI is becoming noticeably faster, often surprising AI scientists.
For an overview of how AI works and its history, please read the entry on Artificial Intelligence in the companion book, A Chronicle of Ideas: A Guide for Superheroines (and Superheroes).
In the novel, Danielle uses artificial intelligence in her computer programming, in her effort to combat tyranny in Libya, to analyze crowd images and provide security during her flash mob celebrations, and to capture Osama bin Laden. You can follow her example by using AI in your daily life now and helping bring about the Singularity in the future.
To start on this path, you should make an effort to learn more about artificial intelligence, how people use it, and where it’s going. Here are some explanations you’ll find useful:
- A quick two-minute intro to artificial intelligence2 by Murray Shanahan, a roboticist and futurist who studies the Singularity.
- A more detailed explainer video about artificial intelligence by Frank Chen,3 who studies AI at one of the top investment firms in Silicon Valley.
- 10 interesting applications of AI4 in products that are already available.
- TED Talk by Google star computer scientist Jeff Dean5 on how AI will affect your life in the coming years.
- My own fuller discussion of the Singularity6 and what it means, recorded for the Big Think series.
- Animated video by the Future of Life Institute7 with commentary by AI researcher Max Tegmark on myths versus facts about superhuman AI, or superintelligence.
- A key question about superintelligent AI is whether it might try to destroy humanity. Here are two thoughtful perspectives on this question, from ethicist Sam Harris,8 and risk expert and philosopher Nick Bostrom.9
- My TED Talk on why we shouldn’t look at superhuman AI as a danger that will be separate from humans, but rather an extension of the abilities of our own species, due to the coming merger10 between artificial intelligence with human intelligence.
- Two outstanding blog posts by Tim Urban at Wait But Why (Part I11 and Part II12), explaining in conceptual terms what steps will be necessary for AI to reach superhuman levels. Tim uses amusing stick-figure graphics and does an excellent job explaining complex ideas in ways that are easy for anyone to understand. He also discusses the disagreement I have with some other futurists about how best to think about risks from AI. Here’s Tim’s Talk at Google13 expanding on the ideas in the blog posts.
- Billionaire entrepreneur and innovator Elon Musk has begun early development work on a project to develop one approach to the brain-computer interface that I predicted in The Age of Spiritual Machines (Viking, 1999). Called Neuralink,14 it involves putting a many-lead connector in your brain that can link it with the Internet. Musk’s design will be appropriate in the near term to provide a brain-to-machine interface for people with serious neurologically based disabilities, such as quadriplegia. My own vision is for this interface to be provided by nano-scale machines (that is by medical nanorobots) to provide direct access to almost unlimited knowledge in the Cloud, and enable direct brain-to-brain communication with other people. Wait But Why has a long but highly worthwhile blog post about Neuralink15 and its implications.
One of the most important areas of artificial intelligence is machine learning: the ability of an AI system to learn and to adapt and improve performance based on its own experience. Machine learning is based on rough models of how the brain works and learns by being exposed to many examples of a situation (for example, many images of different categories such as cats and dogs, or many examples of movies in a game). Take a closer look at machine learning here:
- A basic intro to what machine learning is,16 why it is different from other kinds of computer programs, and how it is used.
- This clear explainer video by Android Authority17 digs deeper into what machine learning is and how it works.
- Udacity’s half-hour intro to machine learning,18 including some useful examples that help you see how these concepts are applied in the real world.
- A short video about machine learning19 from the perspective of Google, featuring interviews with the company’s AI engineers.
- HubSpot’s animated video explaining machine learning20 and the idea that AI and humans shouldn’t be seen as in competition against each other, but rather as working together to solve problems.
- The Jeopardy! match between an IBM artificial intelligence system named Watson and the two top human players in the show’s history, Ken Jennings and Brad Rutter.21 Notice that although Watson’s first choice guesses are usually correct, her second and third choices, displayed at the bottom of the screen, are often not even using the right category of the key word. This shows that although the machine learning used to train Watson gets to similar final answers as human intelligence, Watson’s “thought process” is often different from that of a human player.
Machine learning is probably the best way to get started on learning how to apply artificial intelligence yourself. The first step is getting some experience with the basics of computer programming in general. For more information on this, see the entry for “How You Can Be a Danielle and Learn to Program Computers from a Young Age.” There are also some very helpful resources available that can specifically help you build your skills in machine learning:
- An excellent overview from HackerEarth22 on how you can start from scratch and get good at coding machine learning. It includes information on how to learn programming languages that are good for machine learning, especially Python. It also gives you pointers on the specific coding techniques you should master, and walks you through learning how to build your first AI bot. Once you’ve created your first simple artificial intelligence, you’ll better understand the possibilities of more complex AI.
- If you already have experience at computer programming, this detailed guide from Machine Learning Mastery23 can be very useful on how to dive deep into machine learning techniques and develop advanced skills.
- Another advanced tutorial on machine learning,24 which is good for students who have already taken some college-level math.
As your skills improve, you’ll start getting more ideas for ways to use machine learning to solve problems you see in the real world. Fortunately, you don’t have to build those programs from scratch. Google has created a free, open-source library of machine learning AI called TensorFlow.25 TensorFlow has a wide range of pre-built tools you can use, modify, and expand to meet the needs of your project.
Here’s more on neural networks, the primary tool used in deep learning:
- An explainer video about Google Deep Mind,26 the project that built AlphaGo, an AI system that mastered an ancient board game called Go and defeated the world’s best human players. This was an important breakthrough because the usual method of AI game playing (logical analysis of move-countermove sequences) doesn’t work for Go because there are too many possible moves at each point in the game. The Deep Mind team used neural networks to learn subtle pattern-recognition insights. It originally trained the neural nets using transcribed online games and then greatly augmented this data with simulated games of the program playing itself.
- A closer look at neural networks27 in general, how they work, and why they have so many advantages over other approaches to AI.
- A video on how to build a simple neural network in four minutes.28 A good intro for those who already have basic programming knowledge.
To apply neural networks to problem solving, there’s a whole set of related skills that it’s helpful to learn. If you intend to study AI in school, or pursue a career in this area, you can get a great start by taking some online courses:
- A free college-level course on artificial intelligence from MIT.29 Covers topics like reasoning, neural nets, and visual object recognition. And here’s a lecture from another MIT course, Intro to Machine Learning.30
- A highly popular introductory machine learning course from Stanford,31 available for free through Coursera (though for a fee you can get a certificate verifying your mastery of the concepts). Covers a broad range of topics, and includes graded assignments to test your skills.
- Deep learning refers to AI that uses “hidden” layers in its neural networks between the input and output layers. The use of many hidden layers is the key to deep learning being able to make subtle and abstract judgments. If you already have solid coding experience, take a look at this series of five Coursera courses,32 created by Deeplearning.ai in partnership with Nvidia and the Deep Learning Institute. Includes projects and case studies about real-world applications.
- A Udacity course on Artificial Intelligence for Robotics,33 by Georgia Tech. This free advanced course takes about two months to complete and covers AI programming techniques relevant to robots that interact with the real world.
Another great way to develop your skills in artificial intelligence is through competitions and “hackathons.” These are events that bring people together for a short period of intensive collaboration to solve a particular problem or group of problems. Here are options you can look into:
- The FIRST Robotics competition34 is an international high school robotics competition started by legendary inventor Dean Kamen. Teams of students, coaches, and mentors build and program game-playing robots weighing up to 120 pounds. In 2017 there were over 6,000 teams. Kamen created FIRST to encourage young students to become excited about high tech careers. Participation is about equal between boys and girls. Regional competitions lead to a national competition. These competitions have all of the excitement of national sports competitions.
- Battlecode35 is MIT’s annual programming competition, which is open for anyone to enter. People compete to design AI systems that can outperform their rivals at a game called Battlecode. Prizes are over $50,000 dollars. More significant than the money is the credential of winning competitions such as this.
- The AI Games36 are a set of online competitions to build the best possible artificial intelligences to play games like Connect Four and Texas Hold ’Em.
- Vindinium37 is an ongoing AI programming challenge, where players can use the programming language of their choice to win an online fantasy game.
- RobotChallenge38 is an annual robotics competition that promotes innovative artificial intelligence systems. If you’re interested, get a team together with like-minded friends and compete!
- There are also one-off events, like the Global AI Hackathon,39 which brought together programmers, designers, neuroscientists, data scientists, and idea generators of all skill levels, and had them work together for three days in 15 major cities around the world. There are often events like this—see what’s coming up in a place near you, and sign up! For example, New York City had the NYC Artificial Intelligence Hackathon,40 with the satirical theme of creating technologies that “advance the robot apocalypse.” By thinking about those problems carefully and collaboratively, you can help make sure a real robot apocalypse never happens.
- The General AI Challenge41 is co-Sponsored by Microsoft, and includes a range of challenges for anyone who can achieve milestones on the way to human-level artificial intelligence. This is a very high-level challenge, which will give out $5 million in prizes over time.
In the coming years, especially if you focus your career on artificial intelligence, you’ll have to start thinking more seriously about the ethical implications of AI. If you start learning about those questions now, you’ll be able to become a leader in society’s debates about how to use AI safely. Look over these links as a starting point for going deeper on these issues:
- My thoughts at SingularityHub42 on how to keep artificial intelligence “friendly,” through human integration with AI. As our abilities become more heavily augmented by AI, humans will have to place new emphasis on how to limit conflict between people. Democracy, liberty, and compassion will discourage humans from using superintelligent AI in destructive ways.
- When you code artificial intelligence, your biases and worldview can shape the way the AI works. The data you give an AI to train its machine learning algorithms will affect the lessons the AI learns. So it is important to keep this risk in mind and take active steps to prevent negative human biases like prejudice from being reflected in the AI we create. This short video by Google explains.43
- An animated explainer by WhyFuture44 about the road to superintelligence and the ethical questions we’ll face in trying to steer AI toward positive outcomes for humanity.
- This engaging and informative video about AI personhood by Hank Green from the Crash Course45 series on philosophy. Personhood is the question of whether a human-level AI could have true consciousness like humans do, and whether it might be entitled to rights like humans have. For example, if an AI tells us that it is conscious and deserves freedom, should we believe it? This is a theme of my movie The Singularity is Near,46 which is a companion film to my book of the same name.
- OpenAI,47 a project co-founded by Elon Musk, which facilitates international collaboration toward the goal of achieving friendly superintelligence that integrates well with humanity. OpenAI takes inspiration from the (mostly successful) international efforts to regulate nuclear energy and prevent such potentially destructive power from falling into the hands of terrorists.
- MIRI,48 the Machine Intelligence Research Institute focuses on figuring out how to design AI in ways that can make it naturally safer. For example, they study how advanced AI can be made more transparent so programmers can better understand why it makes the decisions it does. That way, problems can be identified and fixed before they get out of control. You might consider donating to MIRI, or starting a fundraiser with friends who also care about the future of AI.
As you become more proficient at building and applying artificial intelligence, think of how it might be able to solve some of the difficult problems in your community and the world. Consider the areas where AI has natural advantages: making fast computations, processing huge amounts of data, following complex logical rules, and recognizing patterns. How could those abilities boost and complement areas where humans still have advantages, like social skills, abstract reasoning, interpreting complex situations, and hand-eye coordination? The innovations you create this way can bring great benefit to society and bring us closer to a beneficial and safe world of integrated AI and human intelligence.