Last week was the RPA and AI Meetup London, superbly organised once again by Bloom Search this was the 3rd of their quarterly events. This time the format changed from the previous presentation format to a panel discussion, hosted by Andrew Burgess, and a panel of experts:
Terry Walby, CEO of Thoughtonomy, Intelligent Automation
Andrew Anderson, CEO of Celeton, AI and Machine Learning
Karolin Nakonz, Executive Partner at IBM, Cognitive Process Automation
Kieran Gilmurray, Director at Pearson, Blockchain and RPA and Process Excellence
Andrew Burgess is an Independent Management Consultant and author of The executive guide to AI. Andrew sees the value from AI as being orders of magnitude greater than what has come before because it augments existing roles. AI is still developing and provides good business benefits when focused on narrow capabilities. Andrew's outlook is positive and sees AI in business as augmenting, not replacing, peoples roles.
Inevitably there are risks associated with AI:
Lack of Transparency - Its difficult to reverse engineer how decisions are made
Unintended Bias - The AI isn't biased but the data used could be
Over dependency - Assuming 100% truth and not keeping the human in the loop
Society and The Future of Work - Hollowing out of the middle classes as knowledge work and manual work still have roles, but the middle roles become automated
But, there are lots of successful AI Case Studies, for example:
Tractable determines if car is write off
Deutsche Bank use voice recognition to evaluate calls for fraud on every call (used to only be a 5% sample)
Google Deepmind evaluated cooling in their data centres and was able to predict consumption needs reducing the cost of cooling by 40%
Kieran Gilmurray from Pearson working to automate back office tasks using RPA. Pearson Education's AI ambition is to augment and scale across the organisation working with Watson to create individual learning journeys for customers, allowing earlier intervention and tailor training content to their individual needs.
Karolin Nakonz from IBM, works for the Cognitive Transformation Centre. Working in the areas of government and healthcare she sees a wide range of opportunities to transform processes, and views robotics and cognitive as complementary. AI is being used to help human beings to look in the right area, or chatbot for self-help people with arthritis,
finding ways to apply AI to help society
Andrew Anderson introduced Celeton as a platform as a service, handling lots of different electronic data streams, unstructured and unpredictable data and processes. There are many different approaches to AI and selecting the right approach for the problem is essential.
Terry Walby introduced Thoughtonomy, founded in 2013 and now operates globally with over 160 customers in 28 countries. The Thoughtonomy solution combines Cloud to provide elasticity of resourse and ability to scale to fluctuating demand, with RPA, interacting with existing system using the interface a user would use, and AI to self-regulate what resources are needed to to deliver business outcomes and manage sources of information that are ambiguous or unstructured.
On the Impact of RPA and AI...
Terry - Efficiency, Productivity, Effectiveness and Consistency. Improve productivity of non-manual tasks to deliver a better business outcomes. Despite the fear of job losses, actually the things that are being automated are the things that get in the way. It's not displacing people, it's making people more effective. Have seen people reduce labour but also offer new services.
Andrew - Retaining the human in the loop and reducing manual effort. Incorporating AI is not instant, it takes time and do not expect 100%. Apply machine learning and NLP to make a person more productive. Virgin Trains for example reduced manual effort by 85%.
Karolin - When faced with the question "what can we do to avoid birds on helicopter platform?" Watson identified previous research on a solution. Data curation and making data easily accessible is one area where AI can make an impact. A similar idea can be applied to tax cases and many other problems involving a large amount of historical information.
Kieran - The importance of the business case, can you use AI, yes, but should you? A previous company investigated using AI models to find solution to affordably cutting grass around pillons, but solution was to put some sheep in the field. AI isn't right for every problem faced!
AB - American Legal system has a process called e-discovery where a SME trains the software on a small % of documents then the AI looks through the other documents. Courts are happy that if you can demonstrate a robust process then they are happy with the outcome.
Andrew - There is a lot of scaremongering, suggesting AI will change the world for the worst. I don't believe in that. Keeping the human in the loop, the Human is essential. It's not a technology you can just "switch on" as some expect, the analogy here is a school leaver who needs to be trained, it will take time and as you gain confidence in their skills you will leave them alone more often. Just like a school leaver, if they train incorrectly, the AI will learn incorrectly.
Karolin - Watson trained on data that hospital and doctors thought trustworthy. In early days it gave a % rating but this was taken away because it implies a level of accuracy that isn't there, now it uses High/Medium/Low confidence. Also Watson Oncology will provide evidence to why the conclusion, it will provide the source of data to how the conclusion was was drawn so it's not a "black box" as other solutions.
Where are you focused?
AB - When faced with the question,"I need some AI in my business, what do I do?" it is necessary to turn the question around, what is the strategy? what is your business challenge? what are you trying to solve?
Andrew - focus on a large company, ready to disrupt their sector and have a demanding consumer. The area we are working on is the B2C companies that have a demanding customers. It's the customer that is creating the challenge.
Terry - Thoughtonomy do not focus on a specific industry but rather provide a general purpose platform for digital labour. Customers layer their process on the platform to deliver the required (better) outcomes. Partners provide the local business expertise. Biggest customer 450K employees and smallest is a startup with 4 members of staff.
Andrew - Data is critical. Platform learns from historical data and enquiries. Very few companies have learnable data. So have to use the data coming in each day and learn from that.
Kieran - its never going to be 100% perfect at the start.
Terry - AI needs data to make decisions and data is a by product of automation, combining the two technologies allow you to automate as-is then generate data to improve.
One bit of advice if someone is starting out on their AI journey...
Terry - Start with the business problem. Technology emerging all the time, listen to the market to understand what can be created (example customers didn't know they needed an iPad).
Karolin - Unrealistic expectations, lots of un-thinking needs to be done! Do not pick too complex an area, if vision is too broad then there is a risk it will not deliver value for a long time.
Many clients want a chatbot to answer everything! Start with the questions that come frequently that you can train really well, then find ways to deal with the long tail of complex questions. Implement a strategy to deal with the long tail.
Andrew - Dislike term AI! It's about understanding the problem and then applying right tools. Keep it simple. Like hiring a school leaver, build trust and confidence over time. Failing to do so, confidence will be destroyed and you wont touch it again.
Kieran - The right business problem, the right size and look ahead, is there more than one problem to solve once you've invested the money. Communicate, Advocate and sell the story.
Are there data sets AI shouldn't have access to?
Karolin - Do not exclude any domain up front, there is no universal truth, oncology data set was a specific hospital with a specific view. Careful in being clear in who has trained the systems with which philosophy.
AB - Using data to identify children at risk - the data is public services data, which typically is only used by poorer people, the data is not there for middle and upper classes, but this does not mean their children are any less at risk. The challenge is getting the right data.
Kieran - Augment - the example of cancer data where individually the AI and the doctor both achieved a lower result than when both working together. The human ethical lens is important. If you can identify children at risk or cancer then we should.
Examples of doing something more useful with the people rather than cost cutting...
Terry - rarely has it been the case in businesses, companies are more often constrained by resources from doing what they would like to do. New roles are being created to develop processes, understand business workflows and develop robots; these jobs didn't exist 5 years ago
Andrew - Many examples, but no job losses, if you employ a person they can be more productive and do more. New opportunities within the field is increasing the number of people employed.
Karolin - at a recent RBS presentation they thought that introducing messaging would reduce the number of calls into the call centre and reduce staff. They found however that because the method of contact was easier, more people use it and the number of transactions goes up. Automation was the only way to meet this customer demand.
Rapheal Qusestion - What problems best suited to ML
Andrew - There are lots of AI's, lots NLP approaches, lots of ML approaches - fitting the rights types of solution to the problem is a key skill.
Alan Question - What skillsets are needed in the future...
Andrew - Mix of business analyst, data scientist, business process, workflows and they're not the developer! Skills that have background of different kinds of AI techniques plus good coding!
Terry - The client will build the process flows into the platform themselves. Self sufficiency for a business user, not a developer. Goal is democratise access to AI. 3-4 day learning to how to use. Working with a major retailer it is ex customer service staff who are developing the automation. Do not underestimate the ability of people who are not educated in the traditional sense if we can make it intuitive.
Karolin - Its more a governance and knowledge management role once it's all one workforce.
Rob Question - On the role of government for regulation...
Karolin - There is a role for government. If it now fundamentally the workforce, then taxing, why not? The most important aspect is ethics and governments shouldn't rely on self governance.
Kieran - Government involvement might inhibit innovation; plus why tax robots? why not tax apps.
Andrew - Grateful to the government for tax credits in all other points should stay away.
Terry - Can't tax technology. To do so is damaging.
Key Take Aways:
Start with the Business Problem
AI is used for Augmenting roles, not replacing them
When using AI, Focus on a Narrow Problem
Use the right tools and techniques to solve the right problems
On Impact: Efficiency, Productivity, Effectiveness and Consistency
Communicate, Advocate and sell the story
Customers are driving increased demand
The one low point of the evening was they ran out of beer!