Never mind the robots, here’s the real AI

Welcome to this Ko call video blog. My name is Matthias I'm lead advisor at Ko AOL Analyst. I want to talk to you about the myths and the value of artificial intelligence when it comes to its use in a business context, the title of this short presentation is nevermind the robots. Here's the real AI, how to let machine learning add value to your organization. Why nevermind the robots. There is a, an image that is closely connected to the term artificial intelligence and machine learning. And that is of course the picture of robots.

So in the media and the software industry and the press, every time when artificial intelligence or machine learning is mentioned, it is always illustrated with a picture like that one that is of course, coined by visionaries and futurists. That raise expectations that cannot be met today and might not even be met anytime at all, software vendors are using this terminology to describe the capabilities of their software.

And we can think of that often as just some vendor induced type, because products and services need to have some AI in their description, in their feature list, just to make sure that it has this image of intelligence and machine learning built in. So you need to have some AI built in. And when it comes to the common perception in the public of artificial intelligence and machine learning, then this is often considered to be something that is sci-fi like miracle technology often with a negative connotation that it's dangerous, but powerful a threat to our jobs.

And of course, out of control a bit terminology to start with general AI, of course again, here, robot again, what is general AI general? AI is when we come to the definition and the literature, it's the ability of a machine to intuitively react to situations that it has not been trained to handle in an intelligent way, in a human way. That is the notion of general AI. So really a system that acts like described, and the main characteristic of general AI is it does not exist. On the other hand, the public is interested in having AI, even if it is not AI and users love artificial intelligence.

And here are some examples, the natural language processing of Alexa, the recommendation system of Spotify translation provided by picture recognition in Google search, intelligent automation. There are lots of mechanisms in place where AI and machine learning already have shown their value and end users love AI. The senior management levels in organizations love machine learning because it is perceived as these strategic technology, which is capable of improving existing operations and even of opening new lines of business, whether this is true or not.

So the CX O's love machine learning and AI sells many products as mentioned, have machine learning and AI in their names or specs. And this comes with the notion of efficiency and intelligence, which even leads to pseudo AI or fake AI. So this now with AI label often conceals a high proportion of conventional technologies, such as pattern recognition, analytics and statistics, which are really capable, but are not AI, especially when it's just an AI rebrand. There are many products and services around, which are entitled AI, but aren't, and if you look to the left, these are not robots.

These are bottles. So we are talking about a hype and two popular examples when it comes to the use of AI in, in air quotes, real life environments are the two given ones. The first is AI driven intellectual property management. But this is a very good example where AI is not yet in the position to actually implement something like that, because feels that rely on IP protections could influence the algorithm designs to favor certain innovations and public actors could increase the special to receiving IP protection, to destabilize the monopolistic control. That patterns assure.

So, although this is an example that is often mentioned, it is not really in existence and the same is true for autonomous vehicles. Many organizations are working on that, but we must go far beyond the car. The road infrastructure would need to be equipped with a more complete network of IOT sensors to provide enough data to operate safely. And that is just not the case. And we are still talking about ethics here. So the legal responsibility, the, if ethical responsibility of programmers manufacturers, IOT device networks, and passengers would need to be clearly defined.

And as long as this is not the case, we are just not yet there. So these two examples often used to describe the bright new future of AI machine learning is just not yet there, but that does not mean that there is no value. What we like to think of is the so-called cognitive enterprise. These enterprises are striving for innovation, so to achieve and implement the agility required and the culture to change the way organizations work and deliver value to a more modern approach.

We have a bunch of technologies around that can be used with AI and machine learning, being two of them, but not all of them. So there's blockchain, there's virtual reality. There's augmented reality, of course, the internet of things, new telecommunications networks. And for example, 3d printing. And this list of course, is not complete. These cognitive enterprises are implementing their systems in hybrid infrastructure.

And this means that they are moving from on-premises traditional data centers towards the cloud and provide hybrid, secure, scalable, and standardized environments between multi-cloud and on-premises of course, data is one of the core aspects to consider. It's open data. It's public data, it's subscribed data. It's your own enterprise proprietary data. Pre-processed data structured and unstructured heterogeneous data.

All of this is fueling this cognitive enterprise and that all needs to applications, to platforms, to systems available via application programming, interfaces APIs, and these applications are legacy. They are adaptive, they are cloud native. They are just accessible via APIs and all these building blocks together form something that we consider to be the cognitive enterprise and enterprises is capable of using these new technologies and combine all these six aspects into one. We have learned that AI in its pure form general AI is not yet there. So what is different when AI then really works?

Here's the real AI, and that is machine learning. When we want to talk about actual value that AI can provide to a business, it's usually narrow AI trained AI machine learning. So it's codified experience codified through training. And that process enables programs to independently complete a task that it has been trained to do you need to have the right task. So it needs to be applicable so that you address specific repetitive tasks that aggregate complex data that is where machine learning really can show its value and value is the desired effects.

The business value, if it really Actually adds value to your individual unique business, or if it can leverage significant cost savings or it can achieve higher customization and process so that you can make more specific variants of a standard.

So what our characteristics of successful AI use cases, they are successful when they have access to historical data, through training for reference as part of the improvement, when there is a constant flow of real time data, so that the AI can actually react to a critical mass of input for operating and for potential self-improvement constant flow of real-time data. Ideally there is a repetitive task to be completed. As I've mentioned, unlock cost savings through automation. Then AI machine learning is actually perfect.

And when there's only a limited scope of control, and that means that the organization can operate in a contained system without significant interface with other influential actors. So let's take these four characteristics and compare them to five concrete use cases for the use of narrow AI, starting with predictive maintenance and manufacturing. The challenge is easy. It's when needs a device that is installed somewhere else. When does it require maintenance?

And when given the appropriate input such as installation date load, running time and standard maintenance timelines, and AI can actually make the adequate decisions. So the downtime can be factored into production planning, and ultimately can be reduced by having the appropriate parts at hand. So we have historical data, we have realtime data, we have a repetitive task, and we have a limited scope of control, limited to own operations. Second example, network coordination in energy production.

We are talking about the intelligent energy grid and this intelligent energy grid is dependent on inconsistent sources. So we have to have predictive algorithms that are capable of estimating future low supply and automate the support from, for example, passive, renewable energy at the edges of a grid network. So again, we have historical data to compare with. We have real time data. We have a repetitive ongoing task.

The scope of control is a bit bigger, but nevertheless limited because it includes edge devices that might be private property fraud detection in finance fraud in banking is increasingly common. As customers rely more exclusively on just digital online interfaces. So fraud detection can be supported through algorithmic verification of new accounts or account activity. That again, leverages historical data to identify the wanted from the unwanted. It uses real time data to compare to historical data. It's an ongoing repetitive task. And the scope of control is limited.

Although it interfaces with other members of a transaction fourth example, employees support in HR, employees of medium, large corporations can feel removed from helpful resources. So they need some personal interfaces to an HR department. If employees do not properly understand their benefits and repeatedly need help to take advantage of their opportunities and AI and an AI conversational interface can be of help here.

So it can be used to improve the accessibility to information about benefits about sick leave rights and special policies without having to track down the appropriate party and individual file. So again, we have of course, historical data here. We have real time data of the employees interacting with an HR department or the conversational interface. It isn't repetitive task because it will happen again and again, and it will be limited to own operations. Fifth example is predictive demand in supply chain management.

So the idea is that predictive demand algorithms allow for cost savings by decreasing delivery times and leveraging strategic warehouse locations. And that can be achieved by aggregating eCommerce data and contextual indicators for training and AI to predict short term consumer demand.

So again, we have historical data about previous purchases. We have real time data that can be used for understanding how demand will develop. It is a repetitive task because it is an ongoing challenge. And the scope of control again, is limited to own operations. We think of AI as an ecosystem. So the artificial intelligent network of the future is building upon data capabilities, platforms, and services. So this is really some kind of pyramid starting with AI data. This is the foundation sharing data, open data, public data.

There are platforms available where individual services, baseline capabilities are available just to use them via APIs. And there are many of them, they are built into AWS. They are built into Azure. You can just gather your services from there. And this is the next level, actually AI services that are available via platforms, but which Are much bigger and work together interact. For example, smart cities will combine vehicle data. As we've mentioned before, these are comprised of several AI solutions and you combine it with traffic data, smart energy management data, and much more.

So you get from the data via other platforms to services, to solutions and networks. Finally, for takeaways, how to let narrow AI, how to let machine learning, add value to organizations. So this is something as a list of yeah, as a list of takeaways for you to take home. First of all, define the scope of control, determine the scope of control you have, because this is critical to the AI success within your organization. You need to focus on use cases that can be contained within a company's internal network, where it can mitigate risk and establish measurable benefits.

Of course, such an AI solution must be providing actual business benefits. So either cost savings or added value or both balance cost savings with added value. So achieve the maximum benefit from your AI solution beyond cost savings. This includes added value through improved efficiency, higher resiliency, or better communication. Third takeaway, understand your ecosystem. This is again the data aspect.

So if you collect more varied data from more sources that will bring the company into contact with other actors, for example, sensors data from machines, customer information, chatbot conversations, understand your ecosystem and where AI actually makes sense. Where is the data that you can use to benefit from? And that also means that you need to understand how data flows, make sure that you are prepared with adequate data management strategies, that you understand how your data flows, because this is important for the success of AI solutions in your business environment and data flows.

Just as a final thought, here is increasingly concerned with data privacy, with the protection of personal data, but also with the protection of intellectual property, your customer data, your consumer data, make sure that understanding how data flows also means what kind of data it is and how it needs to pre protect it when it comes to new use cases within AI. Thank you for your time for listening to this presentation. I hope this was valuable to you. Thank you very much.