Digging into Artificial Intelligence

Abstract: Is AI ready for prime time? What steps can #CIOs take to bring it into their companies? And what will AI’s impact be?

In the early 90s, I went to first AI Congress which was held in Portland Oregon. I remember that every major company was represented at this event. Later, I found it interesting that the event took place in Portland. The land where Lewis and Clark went looking for a North West passage. Just like the failure of Lewis and Clark to find a Northwest passage, AI failed, at this time, to find general-purpose technologies and uses cases. With notable exceptions, AI then faded into the background. However, Tom Davenport and the other authors of HBR’s Artificial Intelligence insist today is different. They assert that AI is ensconced and starting to bring about exciting business change. Given this, it is a good time for CIOs and other business leaders to dig into Artificial Intelligence.

Why AI Now?

Although the reasons are numerous, AI is actually making work easier. AI machine learning makes it easier to answer business questions. In particular, machine learning is really good at processing vast amounts of information and answering questions that were hard to answer any other way.

Where is AI Today?

It is still early but the HBR authors believe a revolution is coming. AI, they say, has become a key disruptive technology. With this said, it is time to tune out noise and understand AI business implications. Accenture’s Reworking the Revolution Survey has found that 75% of companies are accelerating investments in AI and other intelligent technologies. Even more interesting is that 72% are doing so to respond to a competitive imperative. Of the roughly 25% of legacy companies that have succeeded in deploying AI, some of them have hundreds if not thousands of products. And firms using AI most aggressively turn out to be large businesses with extremely large data sets. McKinsey claims that AI will create $13T in GDP by 2030. This value will be generated, according to the authors, through better business decisions, improved operational efficiency, and enhanced products. AI will increase revenue, reduce costs, and enable the launching new businesses/value propositions.

What has Changed in AI?

The authors suggest that AI is standardizing with the emergence of machine learning—machine learning is fundamentally different than the software that proceed it. In machine learning, the machine learns from example (unsupervised learning) rather than being explicitly programmed for a particular outcome. Programming is deterministic in that it codifies existing knowledge and procedures.

Some of the biggest advances in AI in the last couple of years have come in the areas of perception and cognition. The error rate for voice recognition while still far from perfect has dropped to 4.9%. I need someone to speak to Google because it and I are not that accurate. Meanwhile, image recognition has dramatically improved. The results shown on 60 minutes are year ago for the what the Chinese are doing are simply amazing and scary at the same time.

Successful AI systems today deploy training sets of thousands if not millions of examples. With big data, this is possible. Speed of improvement has accelerated rapidly based on very large, deep neural nets Additionally, AI skills are spreading quickly. Nationwide’s CDO, Jim Tyo, has used internal and community college training to take traditional programmers and give them advanced analytics skills.

Time to Get Started

The authors claim that being late is risky. They believe it is important to build internal business capabilities for AI now. They suggest that systems developed by vendors or consultants will have little value in the end to the business. Another part of this involves learning how to transition from pilots to production.

One of the problems for followers is that the conventional wisdom is that companies are sitting on troves of data that can be readily transformed into competitive advantage. While they may be sitting on troves of data, if the data isn’t already useful, an organizations ability to capture value from the data quickly will be limited. With good data, fully trained AI models must be fed ongoing operational data to make predictions that have business value.

The authors believe fast followers will lose—late adopters will never catch up. Opportunity for companies to differentiate and defend their business is now. Fast followers are missing an opportunity to learn. The time is now to experiment will AI and build internal capabilities. It takes time to integrate AI capabilities into the business functions. Part of this involves having the time to experiment with customer engagement.

Is Facebook an Example for Legacy Businesses to Follow?

Facebook has been at the leading edge of many advanced technologies. AI is no different. Facebook’s machine learning group has grown to several hundred employees and is running today several thousand experiments. Their leader stresses that it is important to know first the business challenges the organization faces. This is good advice for CIOs, CDOs, and analytic leaders. Facebook group leader stresses that there need to be a business outcome from machine learning. At Facebook, they look at AI as having one of three horizons—1) existing products; 2) advanced machine learning; and 3) three years out from product delivery. With this said, the most fascinating thinking shared had to do with when to revise an algorithm/model and why AI shouldn’t be used for every data or prediction problem.

Succeeding at AI

The authors say their research says that highly ambitious moon-shot projects have tended to be unsuccessful or slow to arrive. For this reason, they suggest that organizations start small and customize what they are doing to their organization’s business context. They stress that it is important to choose projects that can yield results in 6-12 months. Most CIOs that I have talk to in the #CIOChat would agree with this approach.

Additionally, the authors believe that it makes sense to use external partners to bring expertise into the organization faster. As well, they stress that analytics and business leaders shouldn’t chose projects simply because a sources system have lots of data. It is essential that analytic projects demonstrate business value as soon as possible. Part of getting results involves doing what CIOs are used to doing—appointing a leader, building a small team, and communicating the value when it is created. With this said, the authors suggest it is important not to set unrealistic expectations. A part of this involves asking the following questions of a potential AI project:

1)     Does it provide a quick win?

2)     Will its results be too trivial?

3)     Is it too unwieldy?

4)     Are there credible partners?

5)     Can the milestones be hit?

AI’s Impact on Employment

The authors say that AI’s impact on employment is uncertain, but jobs will certainly change. They believe that it is possible for machines to take on many tasks that once required a person to do them. It is possible that half of all jobs in the US economy could be made obsolete. However, machines will at the very same time create jobs including new categories of jobs. HR surveys suggest that ? of workers will be in jobs that do not even exist today in five years’ time.

In the end, intelligent enterprises will have all their processes digitalized, decisions data driven, and machines doing the heavy lifting. They will also, according to the authors of “Designed for Digital”, be building digital offerings that transform business model. For the authors of HBR’s AI, the question is what kinds of new skills are needed for AI/digital businesses? How should they be organized? How do we define the new jobs?

Clearly, smart companies will think beyond labor substitution and cost savings to see a much bigger payoff. They will see AI as eliminating non value-added employee efforts. At the same time, they will see that AI needs humans. Look no further for this than Stich Fix. Stylists work with the results of AI recommendations engine to satisfy customers. The combination is where the power is.

Clearly, companies need to learn how humans and machines work together. Humans will be needed to perform three critical roles—1) train machines; 2) explain the outcomes of machines; 3) sustain the responsible use of machines. At the same time, smart machines will help humans enhance their abilities to amplify their cognitive strengths. This will allow humans to focus on higher level tasks and embodying human skills to extend our physical capabilities.

What are AIs Risks

The authors believe there are risks and limits to AI. These include the following:

1) machines can have hidden biases

2) AI deals with statistical truths

3) ML systems do make errors

After results like Tesla’s failure with completely autonomous self-driving cars, it is important that organizations perform a risk analysis. Organizations need to ask what can possibly go wrong. This will allow the organization to mitigate the damage coming from potential AI failures. Organizations, also, need to control user input into systems and check for racial, gender, age, and another common algorithm bias. Sounds here like many a science fiction movie. But you should analyze how software can fail and have a less smart backup plan. At the same time organization should be ready to have a communications plan if and when things fail.

Companies should take the opportunity to reimagine their business processes, focusing AI to achieve more operational flexibility or speed, greater scale, better decision making, and increased personalization. You should determine the desired transformation in advanced, so expectations are set correctly.

AI is Getting More Emotional

Gartner says by 2022 your personal device will know more about your emotional state than your family. Soon machines will be able to recognize, interpret, process, and simulate human emotions. This is combining facial, voice recognition, and deep learning. A great practical example with be smart IVRs. Another is a Dutch bank rationalizer which stops traders from making irrational decisions by monitoring their stress levels. These two cases imply that there could be an emotional AI value proposition. This process will make machines become less artificial and more intelligent. This is good and scary at the same time.

Parting Remarks

It seems clear that AI has matured from the early 1990s. The time is now to experiment and drive toward some quick wins. Without this, organizations will be left behind. AI clearly is not something that you can be late to or buy into at a later time. The time is now to develop your internal AI chops.