Most People Don’t Realize They Are Using AI Now

Most people do not realize they are using AI now. Actually, they are using a subset of AI, called Machine Learning. Machine learning uses algorithms to parse data that the machine then learns from – and goes forward to make decisions based on what it has learned.

(If you wanted to take it another step, there is deep learning, which is a subset of machine learning, and behind that “thinking like human” AI.)

But machines are not all knowing. The fact that they learn through interaction means they have to interact with something. Often that’s people…and people are not perfect.

The Impact of Machine Learning?

According to Gartner’s December 2017 CIO survey of 3,160 CIOs from 98 countries, 2018 was the year AI Adoption kicked off:

And in their article, “2018 Will Mark the Beginning of AI Democratization”, Gartner notes that CIOs will have to overcome challenges, such as poor or unqualified data, organizations with minimal AI skills, or even just understanding the capabilities of new AI techniques, and where to apply it productively.

So where do you apply AI in the business world? It depends on your business. Some may look to physical AI, like robotics and self-driving cars, some to image and video processing, and some to language processing – a great number of companies start with chatbots. But the trend – and area with the most growth potential – is machine learning.

What is Machine Learning?

The use of “smart” phones is a good example of machine learning. For those of us who have them, we have two features that are commonly used. One “suggests” words to use as we type. As we continue to use it, over time, those suggestions are better tuned to our needs. The “machine” has learned how to anticipate our writing.

The second feature is autocorrect. Autocorrect makes choices on our behalf as to what the “machine” thinks we “should” say.

Why does autocorrect get it wrong so often? Because the data that was originally programmed into the phone might not represent our syntax or communication style. Someone at Apple or Samsung has tried to anticipate how we might speak; too often, however, they were wrong.

When the suggested words and autocorrect work properly on our phones, we can more rapidly fire off the messages needed to run our business, or chat with our friends and loved ones. When it does not work correctly, we find ourselves making awkward or embarrassing comments.

This is the upside and downside of AI. When it works well it makes it easier to complete tasks. When it works poorly it creates difficult situations. In most cases this is a function of the data initially integrated with the AI application and our ability to trust it.

Like most things with IT, bad data in means bad data out which leads to bad communication and bad decisions. The common term for this is GIGO – Garbage In Garbage Out.

As written in previous blogs, one of the pitfalls of business is a lack of discipline around data management. Poor data management with AI will lead to really fast but wrong conclusions. Kate Crawford, Principal Researcher at Microsoft Research (Social Media Collective), and the Co–founder and Director of Research at the AI Now Institute at NYU discussed how these issues have impacted society at the annual AI Now Symposium (check out their AI timeline, it’s impressive). It is not unreasonable to look at the same issues associated with biased data in our own business systems.

Consider whether your best data sets are in your customer data or in operations.

As we move from our historic focus on Information Technology towards a future of Information Management, planning is essential. Data sets must be identified, and the data input or accessed by these systems must be focused on the specific issues, cleared of elements which may create bias.

Final Thoughts

According to Harvard Business Review, while AI applications can cover a full range of functional areas, it will probably follow the money. As such, in retail, Marketing is where many businesses start with AI; in advanced manufacturing, Operations and supply-chain management – but it can be used across the entire enterprise to increase efficiency and cut costs.

What businesses need to do is identify the areas of AI that are applicable to their business. All companies, if looking to invest in AI with hopes of gaining a competitive edge, need an AI strategy…and good data.

PriceWaterhouseCoopers, in their 2019 AI Predictions, list facets of responsible AI that are worth sharing:

1. Fairness: We need to address bias when using AI, and minimize bias in our data and AI models.
2. Interpretability: Can you explain how an AI model makes decisions and ensure those decisions are accurate?
3. Robustness and security: How secure is your AI system and can you rely on its performance?
4. Governance: Who is accountable for your AI systems and are the proper controls in place?
5. System ethics: Compliance and regulations must be accounted for, and companies must consider how system ethics will impact employees and customers.

In their priority list, PwC also suggests, and I agree, “Don’t Shoot for the Moon.” If done right, developing an AI model for one specific task can enhance an existing process or solve a well-defined business problem, while simultaneously creating the potential to scale to other parts of your organization. Shoot for that.