7 AI trends to watch in 2017
From tools, to research, to ethics, Ben Lorica looks at what’s in store for artificial intelligence in 2017.
2016 saw tremendous innovation, lots of AI investment in both big companies and startups, and more than a little hype. What will 2017 bring?
1. Democratization of tools will enable more companies to try AI technologies.
A recent Forrester survey of business and technology professionals found that 58% of them are researching AI, but only 12% are using AI systems. This is partially because applied AI applications are only now starting to be realized, but it’s also because right now AI is hard. It requires very specialized skills and a develop-it-yourself attitude.
But frameworks like Facebook’s Wit.ai and Howdy’s Slack bot are competing to become the Visual Basic of AI, promising point-and-click development of intelligent conversational interfaces to relatively unsophisticated developers. Tools like Bonsai, Keras, and TensorFlow (if you don’t mind coding) simplify the implementation of deep learning models. And cloud platforms like Google’s APIs and Microsoft Azure allow you to create intelligent apps without having to worry about setting up and maintaining accompanying infrastructure.
2. We’ll see many more targeted AI systems.
We’re not expecting large, general purpose AI systems—yet. But we do anticipate an explosion in specific, highly targeted AI systems, such as:
- Robotics: personal, industrial, and retail
- Autonomous vehicles (cars, drones, etc.)
- Bots: CRM, consumer (such as Amazon Echo), and personal assistants
- Industry-specific AI for finance, health, security, and retail
3. The economic impact of increased automation will be the subject of discussion.
Expect to hear (a little) less about malevolent AI taking over the world and more about the economic impacts of AI. Concerns about AI stealing jobs are nothing new, of course, but we anticipate deeper, more nuanced conversations on what AI will mean economically.
4. In an attention economy, systems that help overcome information overload will become more sophisticated.
We’re noticing (and applauding) some very interesting developments in AI that help parse information to overcome information overload, especially in the areas of:
- Natural language understanding
- Structured data extraction (from “dark data” to “structured information”)
- Information cartography
- Automatic summarization (text, video, and audio)
5. AI researchers will continue to tackle (and sometimes solve) fundamental problems.
In 1967, Marvin Minsky, cofounder of the Massachusetts Institute of Technology’s AI laboratory, said, “Within a generation the problem of creating ‘artificial intelligence’ will substantially be solved.” Prescient or just wrong? We don’t know yet, but there are fundamental problems still to be addressed. Still, progress is being made. Here are a few examples:
- Ongoing research into building blocks, including natural language understanding, vision, speech, reinforcement learning information extraction, and hardware system optimization (including for cost efficiency) for AI workloads.
- Systems with improved attention and memory that will be able to tackle more than one problem or more complex problems, such as inference and reasoning—for example, DeepMind’s recent advances in differentiable neural computers.
- Algorithms that require less labeled data; unsupervised and semi-supervised learning.
- Systems that are inspired by some core aspects of human intelligence, including intuitive physics and psychology, rapid model building, and causality.
- AI systems for building robust and reliable software. Peter Norvig gave a fascinating talk on this at the O’Reilly AI Conference.
6. Human-machine interaction will become richer.
In machine intelligence, there is a spectrum that ranges from pure machine intelligence to human augmentation. Advances in emotional intelligence and detection and in human-assisted solutions will allow for a richer interaction between human and machine intelligence.
7. Expect continued attention to issues of ethics and privacy.
Most AI systems are black boxes—and immensely complex. The risks related to AI ethics and privacy are real and will require a great deal of thought (and some consensus). These problems won’t be solved in 2017, but we expect to see some headway.