2016 was a busy year for the emergence of AI, but 2017 is shaping up to be an even busier year of accelerated adoption for AI. Here are our top 7 AI predictions for 2017.
Prediction #1: AI means Augmented Intelligence
For too long the thrust of AI has been to replace humans. For the last 30 years the media has loved to portray AI as meaning ‘Machines are better than Humans’; whether it’s Arnold Schwarzenegger in the Terminator or Alicia Vikander in Ex Machina. We think this has incorrectly framed the adoption of AI within businesses. It has set unrealistic expectations for how AI will perform, as well as created scaremongering about the large scale loss of jobs.
A better framing is realizing that Machines and Humans have complementary capabilities. That, yes, while certain automated jobs may be lost, other more interesting jobs will be created. Machines are great at handling large scale structured computation. Better than humans. But, humans are great at discerning meaning and context. Better than machines. AI is about blending these respective strengths. This can be achieved by deploying Machine Learning with Human-in-the-loop.
So 2017 will be the year that AI means Augmented rather than Artificial Intelligence.
Prediction #2: Machine Learning wars intensify
We enter 2017 with the 5 leading enterprise technology companies – Amazon, Google, IBM, Microsoft, Salesforce – having announced or shipped machine learning cloud offerings. We entered 2016 with only 2 of the 5 leading enterprise technology companies – IBM, Microsoft – having shipped machine learning cloud offerings. With over a trillion dollars of market capitalization focused on the Machine Learning market, the level of competition is intensifying with a chasing pack that includes Baidu, HP, Intel, and NVIDIA.
Each of the 5 leading companies is pursuing a differentiated strategy. Amazon is looking to leverage their AWS leadership. Google is focused on developers and massive scale with silicon level investment. IBM is trying to hide the complexity from their customers with their services capability and winning the mainstream media war with their ads. Microsoft is trying to democratize machine learning in the same way Excel democratized calculations in the workplace. Salesforce is embedding Einstein within with their Sales, Marketing and Service Clouds as well as offering it as a platform for developers.
So 2017 will be the year we see an acceleration in adoption of AI within the enterprise driven by this increased level of go to market investment by the big 5 technology titans.
Prediction #3: Shift from making perfect algorithms to making imperfect algorithms work
Much of the industry focus and media coverage within AI over the past couple of years has been around the question of “who has the best algorithm?”. Google, Microsoft and Facebook- with their deep pockets and research focus- have each contributed to significant advances in image and voice recognition. For example a few months ago Microsoft Research announced their conversational speech recognition technology had reached human parity.
But our own empirical evidence suggests that in the real world businesses are not better off trying to create the perfect algorithm since investment in the algorithm is one of diminishing returns. Rather, as our founder Lukas Biewald artfully argued in a recent Computerworld article, the great challenge for the next decade in machine learning is how to make imperfect algorithms work in the real human world.
So rather than trying to take a model from 80% accuracy to 90% accuracy, businesses should focus on the workflow both pre- and post-algorithm. How can you collect 2x the training data? How can you handle the low confidence predictions (hint: human-in-the-loop)? Typically investments in the business process both pre- and post-algorithm yield increasing returns to investment.
So 2017 will be the year smart executives shift the question from “how do we make the algorithm better?” to “how do we combine human and machine intelligence to improve this business process?”.
Prediction #4: Budgets for Training Data explode
For too long corporate IT budgets have had 2 main line items: people and technology. So they have budgeted for employees and contractors and software. But given the overwhelming evidence of the value of business data (Uber, Facebook, Alibaba, and Airbnb have each dominated the transportation, media, retail, and lodging categories respectively by owning and curating data rather than physical assets) it’s time that there is a 3rd category: data.
Specifically within the realm of machine learning, CIOs need to budget for training data. Buying a machine learning cloud service without budgeting for training data is like buying a car without budgeting for gas. You’ve just purchased an asset with no value. In economic terms, training data is a complement to machine learning. With the cost of machine learning set to fall (see prediction #3 above), then economics tells us that the value of the training data will rise. When something increases its value it requires management attention. Budgeting is the highest form of management attention possible within a corporate environment. If you don’t budget for it and allocate capital, then it will be ignored.
So 2017 will be the year smart CIOs build budgets to automate business processes which have 3 categories: people, technology, and data.
Prediction #5: Data Scientists will finally be able to focus on Machine Learning
The role of data scientist within a business has been around for less than decade. In 2008, DJ Patil and Jeff Hammerbacher used the term “data scientist” to define their jobs at LinkedIn and Facebook respectively. Since then we have had Harvard Business Review crown the data scientist role as the sexiest job of the 21st century and now over 200,000 people globally self identify as data scientists according to LinkedIn.
But the reality of being a data scientist within a company has not been as glamorous as the perception. For starters, data scientists spend most of their time doing the thing they enjoy doing the least. They spend over 80% of their time collecting, cleaning, and labeling data. They do this not because they enjoy the task of being a data janitor, but because their success as a data scientist depends on the quantity and quality of the data they organize. Until recently, data scientists had three pretty poor choices for creating training data from unstructured text, images, audio or video. First, ask their internal engineering team to stop coding and spend time labeling data. This is a tragic misallocation of capital. Second, use an external cheaper source of labor (such as a BPO) but have little to no control over quality. Third, do it themselves (another gross misallocation of capital).
Now, though, there is a viable fourth option which is better than the other options in terms of quality, cost, time and resources. You can use one of the two established platforms such as CrowdFlower or Amazon Mechanical Turk to apply human intelligence at scale to your specific training data set. The platforms differ in terms of quality, cost, configurability and scale so do your homework to work out which is the right one for you.
So, 2017 will be the year smart data scientists use CrowdFlower or Amazon Mechanical Turk to build their customized training datasets faster and at higher quality so they can focus their scarce time on machine learning rather than being a data janitor.
Prediction #6: Ethics for AI heats up
The rise of artificial intelligence is forcing us to take abstract ethical dilemmas much more seriously because we need to teach algorithms to understand right from wrong. Should a self-driving car risk killing its passenger to save a pedestrian? Should robots make life-or-death decisions about humans at all? We will have to make concrete decisions about what we will leave up to humans and what we will encode into software.
Ethicists have been thinking about this problem for 50 years with Paula Foot first encapsulating the dilemma in the trolley problem. Imagine you see a trolley barreling down the tracks and it’s about to run over five people. The only way to save them is to pull a lever to switch the trolley to a different set of tracks, but if you do that, one person standing on the other tracks will be killed. What should you do? Depending on your defining ethical paradigm, your response may differ from other reasonable people.
The problem is larger than just what needs to be decided. We also need to consider who makes the ethical decision. Will car manufacturers be allowed to code different ethical standards into their self driving algorithms? Will there be government oversights for these ethical standards? The latest tussle between Uber and California Department of Motor Vehicles is just an early skirmish in this coming battle between regulators, government, citizens and businesses.
So 2017 will be the year regulators, government, citizens and businesses engage in intense and messy debate about how ethics are managed in an increasingly AI world.
Prediction #7: AI goes mainstream
The media coverage seems to imply that AI is only the domain of the technology elite who can afford to assemble large teams of machine learning experts and invest at least $100M. Stories that tend to get covered are the futuristic ones such as self-driving cars. Companies such as Google, Tesla, and Uber are investing hundreds of millions of dollars into being the first to market with a driverless car because of a “winner takes all” mentality. This coverage can give the impression that AI is only for the technology elite who are attacking billion-dollar new problems. But that’s a mistake.
Today it is possible to start to apply the 3 components of AI – training data, machine learning, and human-in-the-loop workflows – to one of your business processes for less than $100,000. So if you’re one of the ~26,000 companies in the US with revenues greater than $50M, you can start to apply AI for an investment of 0.2% of your revenue. So AI is no longer the exclusive preserve of the technology elite. It’s for every business.
AI is also for the more mundane existing problems. One of the core imperatives of any business is to understand your customers. This was true in the first markets such as the agora in ancient Greece and the forum in ancient Rome when buying and selling was done in person. It’s still true today even with the explosion of buying and selling on the Internet. Many companies are sitting on a treasure trove of unstructured data from their customers either in email threads or comments on Twitter. AI can be applied to those challenges of categorizing support tickets or understanding sentiment in tweets. So AI is not only for billion dollar “exciting” new problem like driverless cars. AI is also for million dollar existing “boring” problems such as understanding your customers better through support ticket classification or social media sentiment analysis.
So 2017 will be the year smart executives in tens of thousands of mainstream businesses realize that the benefits of applying AI to their customer data business processes is within their grasp.
** Full disclosure: Microsoft is an equity investor in CrowdFlower and a product partner with the co-branded “CrowdFlower AI powered by Microsoft Azure Machine Learning” solution.