The thesis and final project for my Bachelor of Science in industrial engineering was about a speech recognition system (which actually only understood numbers) that I programmed in C++ using neural networks and Hidden Markov Models. As surprising as it may sound, it was actually pretty limited in scope. I trained it using a data set collected from my entire college dorm population, which consisted of approximately 150 students that patiently stopped by my room in their spare time to say a few numbers to my system. The training processing time was kind of exasperating… it took all night to train the system by running Linux on my Compaq 486. Eventually, the system became very accurate in recognizing numbers no matter who said them. At about the same time, Deep Blue beat Garry Kasparov, the chess world champion…. for the first time! The next 3 games, Garry won… the system was not actually “intelligent,” in the same way a calculator is not intelligent, it just followed rules.
Not even in my best dreams, could I have imagined that some 20 years later the same machine learning principles I used in my numbers recognizer system would be at the epicenter of the incredible cognitive revolution we are immersed in today.
The pace of technological change is unprecedented. After looking at the Tensorflow website this past weekend and browsing through all of the readily available assets that the community has created and shared so far, it seems to me that what took several months to build at the time, could easily be done in less than a week today, and could be deployed with a much better performance even on mobile devices, because where else would we want to use it these days? Our civilization seems to have strongly embraced the African proverb of — “If you want to go fast, go alone. If you want to go far, go together” — and that is great news for this endeavor.
For those of us who are inclined to see the glass half full, AI-powered productivity growth is a promise to lift the global economy, by enabling companies to innovate and reach underserved markets more effectively with existing products, and by allowing for the creation of new products and services over the longer term.
To understand what we currently refer to as AI, its capabilities, as well as to separate common myth from reality, it is essential to know that the machine learning type that has been dubbed “deep learning,” is still based on artificial neural networks which are rooted on how the human brain works. Researchers train neural networks by adjusting the activation functions and their respective weights to get the desired outputs. With a single network layer (as I had in my number recognizer) you can only identify simple patterns, but with multiple layers you can find patterns within patterns. Current neural network systems are typically composed of twenty to thirty layers. This heightened level of abstraction is the main reason for significant improvements in machine learning and AI. There are all sorts of AI variations (supervised, unsupervised, reinforced learning, transfer learning, etc), but I won’t get into that here.
Machine learning relies on a bottom-up, data-based approach to recognize patterns. It utilizes the same approach that my children used to learn four different languages, one of immersion rather than grammar memorization. Simply said, instead of coding the
logic to distinguish data, you simply show the system the data and tell it what is and what is not, such that the computer builds the appropriate program.
Today, there are several factors contributing to this AI-boom:
- Availability of massive amounts of data, the “new oil” that fuels this revolution, that come in all shapes and forms (companies, social, sensors, etc)
- Algorithms of growing complexity are widely available, and often open-sourced by a very active AI community.
- Exponential computing capacity, with more efficient graphic processing and tensor units, aggregated in hyper scale cluster and available in the cloud.
This poses infinite opportunities that could only potentially be limited by our narrow imagination. The challenge is therefore not a computational one, but rather one of evolutionary change in the way we think and work. There is no longer a Moore’s Law wall to hit. Any repetitive task for which there is large amounts of data to train the machines, can and will be automated.
So… this is it? Not yet… What does the future have in store for us? Experts say that all of today’s booming AI use cases revolve around applications of what they call “narrow AI,” in which machine-learning techniques are being developed to solve very specific problems. The major challenge is to develop AI that can tackle general problems in much the same way that humans can, but such “general artificial intelligence,” seems to be decades away.
In the meantime, all industries are leveraging this “narrow AI” and its data to gain new insights and create advances in their field. The internet is inundated with use cases for almost every industry, and below are a few examples that are useful to understand how AI is embedded in our daily lives without us even being aware of it:
- Any home assistant or phone speech recognition is powered by AI. Have you noticed how it becomes progressively better at understanding even small kids when there is background noise?
- Have you tried to perform a search in your photo library lately? Nobody tagged your photos with a dog in them and it is all because of AI playing behind scenes again.
- Facial-recognition algorithms, (your phone’s face ID for example) based on deep learning, have an accuracy rate of over 99 percent.
- AI can alert ICU doctors of a patient health’s likelihood to worsen minutes before their vital signs would, giving doctors invaluable extra time to react.
- Google identifies spam emails and translates web pages into over one hundred different languages using AI. As any other AI algorithms, its trial-and-error process improves as more data becomes available, and as a matter of fact, Google has so much data to improve its algorithms that it is hard to distinguish its translations from those of a linguist.
The dream of any marketer is gradually becoming true. AI allows to go from segmentation to micro segmentation by making sense of a lot of company and social data and presenting pricing and promotion options that companies had never seen before. Companies can know their customers better and be very precise about what to offer to them, when to offer it and through what channels (where). Today, retailers routinely use these “next best product to buy” algorithms, and similar use cases will flourish in other industries to enrich customer journeys.
- Analyzing data from sensors. AI is being used to improve business performance through predictive maintenance.
- In logistics, AI can optimize the routing of delivery traffic, improving fuel efficiency and reducing delivery times.
- Data center growth is exploding and is driven by the expansion of cloud providers (hyperscalers); AI can aggregate and analyze data quickly and generate productive outputs, which operators can use to manage density associated with computing, networking, and storage, reducing power consumption, and increasing performance.
- Unsupervised learning asks the machines to look for patterns in the data, with interesting use cases being the identification of cyber-attacks, terrorist threats, or credit card fraud.
- Have you read Michael Lewis’ Moneyball? It describes how the Oakland Athletics used an analytics and evidence-based approach, instead of the judgment of sport scouts, to assemble a winning baseball team. Today, firms are using this approach for hiring, and advances in analytics and AI have significantly improved the power and accuracy of “people analytics.” Using analytics, some companies are complementing or replacing humans in the traditional interview process for recruiting people of just about any skill. Its use is widespread to recruit part time freelancers or match them against company needs in part time employment marketplaces that plague our gig economy. A data-driven approach to spotting talent allows companies to broaden their pool of candidates and source from universities they did not go to before for lack of time. Human resource decisions ranging from recruiting and training to evaluation and retention are increasingly being assisted or driven by data and machine-learning algorithms.
What does this mean to your industry or business? What are the specific opportunities? What’s the value? What are the use cases? While no solution will typically be a “silver bullet” that makes a genuinely transformative impact on its own, each of them will make your model more difficult to copy, and make it better than your competitor’s.
By now, I may have spooked some readers, since the societal implications are apparent. If machines can read X-ray or MRI images as well as or better than radiologists who have years of training and experience, do we need the radiologists?
According to McKinsey, about half of current work activities (not jobs) can technically be automated, and while less than 5 percent of jobs have the potential for full automation, almost 30 percent of tasks in 60 percent of occupations could be computerized.
Even white collar jobs are under the threat of automation. In reality, the distinction is not between manual and cognitive skills or blue-collar vs white-collar work, but rather whether a job has large elements of repetition, and massive amounts of data are available or can be collected to train the algorithms. Jobs that are routinely, repetitive, and predictable can be done by machines better, faster, and cheaper, and will probably be done by machines, sooner or later. In and of itself, this is neither good nor bad, it is a fact.
In my humble opinion, AI can make us more human. Nobody likes doing repetitive tasks. In the same way calculators allowed people to devote their time to more meaningful and complex jobs or ATMs shifted the role of bank tellers from one of simply dispensing cash to one of customer service, AI will not replace human judgment, but will be a major complementary asset that will free humans to focus on tasks that really need them. Companies will be able to serve underserved populations, and focus their expert’s time on the jobs that really require their expertise.
Automation will spur growth in the need for critical thinking, creativity, complex information processing, and social and emotional skills such as communication and empathy. The so called “high touch professions” will probably be on the rise.
While some jobs will be eliminated, others will change or will be created, and hopefully the change will be for the better. A trend everyone seems to agree with is that the pace of change will be proportionate to the wages at least in the United States..
The core competency we need in the future is the ability and desire to learn. AI will maybe help in telling us what we need to learn.
Conclusively, the consultant inside of me suggests a couple of guidelines for the best ways to leverage AI inside organizations:
- As so many things in enterprise life, it boils down to leadership. Create a business-led, top-down, technology road map. The C-suite needs to create and embrace an overarching vision for how technology can enhance the companies’ performance. This will gradually change the culture as well.
- Align business and technology leadership on the sequence of solutions to be developed. Make sure there is a common understanding of technology so the business can “pull” for the services and for the support of technology instead of having IT “pushing” solutions.
Think of what business you are in, or you can be in. Be the blurrier of your industry’s boundaries. Don’t define your competitors narrowly. Define them broadly or others will do that for you.
- Since data is fueling this revolution, capitalize on it, it is the source of your competitive advantage. This will entail having a data strategy and governance model to ensure data is reliable, accessible, and continuously enriched to make it more valuable. Prioritize the data domains that support an initial set of solutions.
- Rethink your IT platform accordingly — have a data platform at the core of your IT ecosystem and a development environment for producing software and analytics code. There is more value in well managed data than in your core business platform, which quite frankly is becoming a commodity (with some business like Allianz open sourcing it) and just another source of valuable data for your data platform.
- Link legacy and digital applications to the data platform through application programming.
- Don’t try to nail it the first time, instead have short and simple iterations as the best way to move forward. Embrace agile and its MVPs.
- Put the measures in place to make sure that as you deploy new technologies you make value-creating adjustments to other areas’ operating models, so your efforts are not isolated but spread across your value chain. Every technology solution should set up a new phase of operational changes.
- Consultants and third parties may help, but they do not scale; hire people with the skills to refine and extract value from this important resource.
- To overcome functional silos, companies with a COE defining approach, methods and tools, and training the wider organization, with agile cells that deliver the use cases inside the business areas, At everis, we have centers in various locations such as Chile, Brazil and Peru, adding value to our clients’ business.
As companies across sectors are increasingly harnessing AI’s power in their operations, think of how you can best leverage your assets, staying true to your DNA, to embrace this cognitive revolution.
Whether you have a lab, a venture, an outpost in Silicon Valley, incubators or accelerators around the world, do internal or external hackathons, or invest along with other VCs to get a peek into what’s next, or all of the previous combined, make sure that you leverage them all to jump onto the AI wagon sooner than later. This journey will take understanding, courage, conviction, and enthusiasm.