AI reality check. Trying to get to grips with the impact of AI?
Here’s some food for (human) thought. Scientists recently fired up the synapses of SpiNNaker, the world’s largest neuromorphic computer. While it’s a major leap in design and performance – able to perform 200 quadrillion actions simultaneously – in terms of thinking capacity it’s still only (just possibly) reaching mouse level. That’s 1,000 times smaller than a human brain.
Boston Dynamics’ somewhat sinister, door-opening dog created a flurry of concern about imminent robot takeover but designers are still teaching it to handle the messy world we live in. Every move has to be programmed. Millions of terabytes of information must be analysed in real time with every glance. Until quantum computing goes mainstream, we’re not in Black Mirror territory. Focusing on the aspirational end of AI – the ‘bot that inhabits our world and is indiscernible from a human, (possessing Artificial General Intelligence, to use official lingo) is not hugely helpful for strategic planning purposes in 2018. However, it is very important to consider more immediately or imminently applicable AI.
So what are we talking about exactly?
Not a silly question as the definition is evolving with the technology. While impact analyses often lump them together, machines/robots are not a subset of AI but an intersecting one. Yesterday’s AI is today’s dumb ‘bot. A ‘dumb’ bot can still be seriously disruptive. A quick peek inside an Amazon or Alibaba distribution centre will confirm that. However, for the purposes of this article, we’re talking about coded intelligence that interprets inputs and responds accordingly, learning and applying what it learns along the way. We’re talking about ‘bots, chatbots and virtual assistants, which can act as internal or external helpdesks, provide sales support, training and education and more. Increasingly these will be conversational AI platforms, using speech rather than typed text. We’re talking about analytics and predictive analytics – processing large amounts of information (including visual and biometric data) to support human action or decision-making and personalised interactions. We’re talking about AI interacting with smart objects, sensors and environments (the Internet of Things) to create automated and customised experiences. It’s also important to think how your products or services might interact with others’ AI. Will they be friends with Siri or Alexa?
The ‘why?’ of your AI
Without purpose, AI is just an expensive gimmick. Before getting carried away with its cutting-edge kudos, decide what contribution it’s going to make, just as you would with any other business innovation. How will it help your organisation or your customers/stakeholders? What will success look like? For example, there are techie metrics to assess how well your chatbot is functioning (activation rate, session info, conversion sentiment and more). However, defining, designing for and measuring performance against big picture goals is key to delivering meaningful outcomes, according to Jen Hyatt, founder of social benefit AI development company Troo.
It’s not an either/or option
Noise around the potential for AI to replace human labour distracts from the value of collaboration between the two. Hyatt talks about the benefits of keeping humans in the loop during AI training to keep the synthetic character in balance, citing Microsoft’s chatbot Tay (which learned hate speech within hours) as an example of self-learning AI gone awry. AI will only be as diverse as the information that feeds it: a bias in data input will increasingly skew its outputs.
In the manufacturing sector – a sitter for automation and smart ‘bots – there is still a place for human instinct. In Rolls Royce’s automated workshop, machinists play an important role in spotting variance in robot behaviour and forging strategies to reduce it. In a service context, it may be some time before Robolawyer turns up in court but AI driven chatbots can provide cheap-and-cheerful advice on simple legal issues. Suitably trained AI is also extremely efficient and accurate at legal research and checking contractual fine print. There are many other sectors where its ever-cheerful and patient flair for FAQs, eye for detail and talent for tedious admin could be of value and free people up for other more creative tasks.
At our November Future of Work event ‘The Digital Human Debate’, Nikhil Ravishankar Chief Digital Officer from Vector commented “For me there is no doubt that a digital human’s role is to humanise the humans by giving us time to be more human. When you look at it from that perspective the use cases become more obvious.” At the same event, Hilary O’Connor (recently appointed Director of Technical Sales at Soul Machines) highlighted the 70% attrition rate from level one call centre jobs: “This shows us that few people really want to do it...so that’s a perfect use case for a digital human...and that frees up the real human who has a creative mind and is much smarter than just following a script on a screen to be lifted up to level two, to more rewarding work, being able to have creative conversations with customers.” AI can also enhance services that require an honest download of human information to work well. Aside from breaking down barriers of language, distance and time zones, ‘bots are non-judgmental so people may interact with them more openly than with a real person. O’Connor commented on how useful this has proved at ANZ where a digital human called Jamie is improving financial understanding by removing the embarrassment factor. People aren’t afraid to ask her the simple questions. “Vulnerability is a key part of what makes us human and digital humans allow us to be more vulnerable.” This does rely on your ‘bot being well designed, trained and transparent in its purpose. With conversational AI becoming commonplace and beautifully nuanced, emotionally-responsive digital humans from companies like FaceMe and Soul Machines leaving Uncanny Valley far behind, there will no room for bad ‘bots.
The right stuff
An AI project is multi-faceted. You need the right computer architecture to support its greedy processing appetite. You need to understand its algorithmic limitations and digital vulnerabilities and have suitable verification and cyber-security in place. You need to consider its impact on your systems and processes, your teams and their training, your stakeholders and their engagement. You can’t just leave it in an IT silo. It should be strategically (and purpose) led and collaboratively built. So while developers are evolving the likes of Sophia, Actroid F and Erica (who, despite their many amazing features, would definitely not get jobs at Westworld) there’s plenty to be getting on with. Explore and develop an AI project with expert input through our Master of Technological Futures. The next intake starts on 5 March 2019.