- In this lecture, we'll talk about the choices you'll face if you want to use cognitive technologies to automate work. For help with this, I'm joined again by Tom Davenport. We're going to go over the four automation choices presented by cognitive technologies, and the two strategy choices. If you are involved in systems design, job or organization design, or strategy, these ideas can help you think through your options, and answer questions like what should machines do, and what should people do, and how should we redesign work and redesign jobs as we implement cognitive technologies? For decades, researchers have studied automation, its impact on productivity, its impact on workers, and how to determine what to automate and to what degree.
Tom, how should we think differently about automation in this era of cognitive technologies? - Well David, some things have changed, and some things haven't. One thing that clearly hasn't changed is that automation benefits are slow to be achieved. I mean, it took with some previous generations of automation in manufacturing a couple of decades before substantially higher productivity levels were achieved, and we should probably expect that that's going to take place here as well.
There were in the early days of manufacturing automation also people who kind of knew how the machines worked and could configure them and make sure that you have the power loom in the right place and so on. So I think we'll need people like that with cognitive technologies as well. The thing that's different though with cognitive technologies is this is about knowledge work generally, and knowledge workers historically have a lot of autonomy in their work. They can kind of screw up the process a bit if they're not happy, as we've described, and so, I think it's really important to bring them along, maybe promise at the beginning that no jobs will be lost because of this cognitive technology to get them to collaborate with you in designing the work.
- That's a great segue, Tom, to the next point, which is that because cognitive technologies make it possible for machines to do things that only people used to do, it's time to take a fresh look at how we should automate and to what degree. So we've developed a framework that can help. It identifies four automation approaches, each with different impacts on workers. The first approach I call replace. With this approach, you automate everything. Think of replacing bank tellers with ATMs, or call center workers with interactive voice response, or IVR, systems.
The second approach I call atomize and automate. Here, you break up jobs into narrow tasks, that's what I mean by atomize, and automate as much as possible. A person does what's left. Factory automation has been like this to a degree, and in the knowledge work domain using machine translation and having professionals clean up the results is another example like this. The third automation approach is called relieve. Here, automation is reserved for tasks that are dull, dirty, or dangerous.
Examples include using robots to perform repetitive factory work, or to find and disarm landmines. A less dramatic example is in banking customer service where voice recognition technology can automatically identify a caller. This relieves the customer service rep of the need to ask the caller for credentials, so the rep can start providing service right away. The final automation approach can be called empower, or as Tom calls it, augment. In this approach to automation the focus is making workers more effective, possibly by automating what wasn't even being done before.
An example of this is IBM Watson for Oncology which aims to recommend cancer treatments to physicians, providing detailed evidence to support its recommendations. This can help doctors make more informed decisions than before. Tom, how do you see this empower or augment automation choice playing out? - Well, I think that's the one that's going to appeal most to knowledge workers. It appears to be the one that's being implemented in key areas like digital marketing, where we've never really done this sort of task before.
Digital marketers have to decide what ad is going into a particular publisher's site, and they typically have a few milliseconds to do it, and they have to do it millions of times a day. That's not something that humans can do, and so, it's a new task, and I think relatively well suited for automation and augmentation. The same with Watson for Oncology. Watson is a question and answer oriented technology and that kind of presumes a human at the beginning to ask the question and a human at the end of the process to take that answer and do something with it, go treat a patient.
So, of course, it all depends on the technologies that you're using, but something like a Q and A oriented approach can be very well suited to augmentation. - So, exactly, and from these examples you can see that cognitive technologies don't have to eliminate jobs. It can also change them by eliminating some work or by performing tasks that weren't being done before. So, this means that when automating work designers and leaders have choices. A question for you Tom is how should leaders make this choice? - Well, I think in general as we've said, automation mostly goes at task and not entire jobs.
So, the replace one for me is not the most popular option. Atomize and then automate I think is far more common, and if you're talking about knowledge workers I think empowerment, or as I call it augmentation, is generally a good strategy. - [David] But we're going to see some of each I think across the world of work. - Right, and if cognitive technologies are going to be as popular as we think they are, we'll see examples of all of these different strategies throughout an organization. What you'll really have is a whole set of processes that are automated or augmented to some degree.
- I want to build on that by pointing out that besides automation choices organizations implementing cognitive technologies face strategic choices as well. They have to choose between a cost strategy and a value strategy. A cost strategy uses technology to reduce costs, especially by reducing labor, but also possibly by reducing errors and rework. A value strategy increases value by using technology to make workers more effective and more productive, or by reassigning labor to higher value work as technology automates lower value work.
Neither the task nor the technology used nor the automation approach dictate whether to follow a cost strategy or a value strategy. This is also a choice for leaders, strategists, and system designers. Where do you think the balance is going to fall among organizations, cost or value focused do you think? - Well David, I think we probably both prefer the value approach, but that's harder. It takes more creativity. It takes more careful thought about how to deliver more value with cognitive technology.
So, I suspect we'll see a fair number of organizations trying out cost oriented approaches. But those are kind of self-limiting. They lead to commoditization. Everybody in the industry adopts the same kinds of cost reducing automation. Then you don't have any advantage anymore. So, I hope they'll come back to value then at least. - [David] Deloitte's study has shown actually that over a long period of time value creation tends to produce higher performance for companies, rather than just focusing on cutting costs.
So, to wrap up, in this lecture we learned that cognitive technologies present four automation choices, and two strategy choices. There's no single right set of choices for organizations to make, but organizations will have to make choices, rather than assume there's only one way to use cognitive technologies.
- Artificial intelligence explained
- Cognitive technologies explained
- Supervised, unsupervised, and reinforcement learning
- Machine learning models and algorithms
- Language, speech, and visual processing
- Business applications of cognitive tech
- The impact of cognitive technologies at work
- Future of cognitive technologies