We’ve taken a snapshot of our IT & Digital Leaders’ panel featured at FST’s Future of Financial Services, Sydney 2023 event, with panellists breaking through the hype surrounding AI technologies, why they believe it could serve as an ideal intermediary – that tech ‘gobbledygook’ translator – between tech teams and the wider business, and why it may kill off productivity-sapping development team meeting once and for all.
Featured speakers:
- Brendan Mills, Group Chief Information Officer, nib Group
- Kasey Kaplan, Chief Product Officer, Beforepay
- Tim Hogarth, Chief Technology Officer, ANZ
Moderated by David Ridge, Principal Solutions Consultant, PagerDuty.
Ridge (Mod): Organisations are increasingly looking to bring AI in-house, serving those practical and more automated use cases we’re seeing today around developer tools, developer assistance, and some of that generative stuff.
Where does that leave the developer, if writing code suddenly isn’t part of their job?
Kaplan (Beforepay): I think the role of the developer, in fact, becomes more important. Generative code assistance can accelerate processes, but someone still needs to understand the architecture of the platform, make sure that they’re building for the future and that the code is doing what it’s supposed to do.
It’d be quite easy for AI to develop code that does something the wrong way.
At the same time, it also creates safeguards enabling you to have more automation and more checks in place to minimise errors. Like everything, it’s about how it’s being used. For some developers who may be doing more redundant work that isn’t high value, those roles might go away. But for the architects and those more skilled engineers, it can accelerate their work and really amplify it.
Mills (nib): I have a similar view. With quality and efficiency gains, I don’t think it necessarily removes the capability or takes away from the existing skill set. I can think of examples – and we’re blessed with not having this problem, but many do still – like really old legacy programming capabilities or pair programming, where AI can help gain efficiencies. But it also has other useful applications, like log analysis and incident management through cyber events and whatever else you might have, where an AI bot can look for patterns within [massive] volumes of data.
As IT leaders, there’s got to be as much inward reflection as well about how we can make our own shops more efficient by leveraging of the capability.
Hogarth (ANZ): There are many opportunities. The basic one is in helping engineers become faster. I had a colleague who said, ‘Writing code from scratch is no longer a job. It’s just a code reviewing problem.’ But there are a lot of people copying and pasting today; here, it’s going to lead to a step change in the efficiency of engineers.
I’ve worked in a number of banks where there’s a large change required for an application that runs internet banking; it can take a new developer four weeks to work out how to compile and just navigate it, and they’d say, ‘Oh don’t touch that, we don’t know how it works! There’s stuff in there that’s really fidgety.’ There’s a whole lot of evidence that these tools are exceptionally good at parsing a local code base and giving you insights and actually saying, for instance, ‘This is redundant code. This is really buggy. This is not written well.’ It’s stuff that we haven’t been able to do before.
Then there are all those opaque applications that were written several decades ago where we’re not entirely sure how they work and not sure of the source code. Reverse engineering [through AI] is suddenly so much more powerful.
And then you start talking about analysing tests, in automating the DevOps story and the whole ITSM [IT Service Management] process. There’s always stuff that we’ve done before using ingenuity when it should, effectively, be using statistical modelling.
This is just incredibly exciting. It’s going to transform so many industries that are in the software development lifecycle.
Ridge (Mod): When we look at AI and AIOps, it oftentimes goes hand-in-hand with automation. Have you looked into bringing additional automation and those kinds of similar machine-in-front-of-a-human elements to the operational side of things?
Hogarth (ANZ): It’s probably speculative at the moment. We’re automating all of the end-to-end systems to make sure they log their events, they’re reported, the systems are recorded, we have a proper up-to-date database of all the applications at the bank, we know what systems they talk to, we know the thresholds when things go badly, what the trigger points are. But stuff still goes wrong! And you’ve got people trawling over all of that.
Now, with AI, you’ve introduced this whole idea where you can simply ask the system, ‘Would you mind comparing this incident with that incident, and what was similar about them? Just trawl the logs and find what was different. Or, say, ‘There are a lot of engineer reports here. Can you explain to me in a paragraph what went wrong?’ You can still use a human to do this, but these tools allow you to do things you haven’t done before.
Instead of trying to define the perfect Splunk dashboard, you can ask, ‘Can you tell me where all of the security incidents were last year and what’s the most likely root cause or the top 10 root causes and what I should do about it?‘ That was unimaginable even nine months ago!
Ridge (Mod): So asking those higher-order questions?
Hogarth (ANZ): So much of what we do inside IT and the banking industry is translation: so, translating a high-level business request into a specification, into technical exercise that goes into code that goes into machine language. It’s only that last bit we’ve actually automated!
Now we’ve got this process: instead of trying to conduct that translation in the least effective human enterprise – the meeting! – where you discuss this, you’ll be able to use this language to do the translation in seconds. This effectively eliminates the meeting, and I think we’d all like to eliminate at least some of our meetings.
Mills (nib): That’s a great example. At the end of the day, it’s that concern over quality time, and where you want to be spending it, particularly when you’re in the middle of an incident.
We’re not playing in this space at the moment. We do see an opportunity, but we’ve been more customer-facing, efficiency- and financial-focused.
However, if you’re in the middle of an incident – be it a cyber event, an outage, or internet banking is down – do you want your cloud engineering manager taking their time translating a summary that you can then on-send to your CRO or to your BCP team or your CEO, or do you want to ask some sort of Generative AI tool to be able to do that for you? That’s a perfect use case.
Kaplan (Beforepay): I’d agree. It allows more people to come into the conversation. When you’re dealing with technology, anything in financial services, it can get very complicated. And when you bring in legal compliance or other areas, they might not understand what you’re talking about.
Being able to synthesise the conversation, other people can ask questions in their own words; it breaks it down in a way that they can understand it and ultimately allows the incident response team to be much more effective.
Mills (nib): It’s a good example where it can be 90 per cent correct, but it’s still good enough to share. For the information you’re trying to transpose, it’s fine.
Hogarth (ANZ): In large organisations, when there’s an incident and there are a lot of people running around and triaging, the most important thing is how quickly you can get back [on track].
You do lose information and translation from the tech team to the legal team, to the comms team, to the customer team. And that can exacerbate the problem dramatically because of the fog of war. Often, just getting it slightly better and cutting down that timeframe is going to make a much better outcome for when you’re in crisis management.
Kaplan (Beforepay): We just went through an audit recently, and I had to spend an hour explaining to our auditors what an ‘API’ was and explaining how our systems work. Through these types of solutions, I wouldn’t have had to do that. They might be able to prompt some of the questions and break them down in terms that non-techs might understand.
Ridge (Mod): It’s interesting about APIs. I’m old enough to remember when it was an initialism and not a business term. Do you think that ‘AI’ has that vibe around it now, where it is now a business term rather than a technology? Where you might hear, ‘Oh, we’re doing AI now!’.
Hogarth (ANZ): It’s frustrating. Consider Arthur C. Clarke’s quote: ‘Any sufficiently advanced technology is indistinguishable from magic’. People are asking whether they can use AI to solve problems that have nothing to do with AI, but everything to do with a decision not to upgrade a system: ‘Can AI help me upgrade that?’ No, let’s just make a decision and get on with it! So, yes, it is an overused word.
The problem with the word ‘artificial’ and the word ‘intelligence’ together is that it personifies something that shouldn’t be! It’s a statistical probability, a model.
It’s essentially a statistical likelihood of something occurring: instead of writing rules, just give me the probabilities of the rules.
If we stop thinking about it as ‘intelligence’ and start thinking about it as a probability machine, it would probably take a lot of the hype out of it. But we’re not going to stop that now. We have to keep talking about it, and we need to educate everyone and realise where it is prescient in its predictions, and where it’s utterly unreliable.
Kaplan (Beforepay): Since the term AI has been thrown around in the last decade, it means different things to different people, and often it’s synonymous with automation. I don’t know that we’ll go back from that; it’s definitely a business term more than having any practical meaning at this stage.
Mills (nib): Tim mentioned education. We’re really focused on that internally at the moment. There is a massive opportunity to work with our business leaders around what AI is, what it isn’t, and how we can apply it.
It’s not, after all, going to make coffee; it’s not going to fill water up at the water cooler. Everyone’s read a lot, there’s a lot of hype. People want to apply it – at least we’re seeing that. But the education piece is really, really key. It’ll help our internal IT, our technology and transformation teams and business leaders almost supercharge their businesses, and determine where they can get value versus where they can’t. But, at the moment, there’s a lot of unknown in the stratosphere out there.
Ridge (Mod): That brings us back to the initial question: Can AI make you a leader in the market or is it simply an assistance tool?
And where do you think we are on the hype curve? Are we on the way down or on the way up?
Mills (nib): We’re on the way up, I’d say.
I’d certainly hate to be a senior leader in an organisation that isn’t really looking at how they can apply this. There is a real danger that organisations get left behind.
Where this ends up in five or 10 years’ time, I’m not sure any of us can actually predict precisely. But I do think there’ll be some people playing catch up, and there’ll be some people that are at the forefront; that’s the way of these technologies. It’s like cloud, mobile technology, digital and probably internet banking back in the day.
Hogarth (ANZ): We’re still rising, because we haven’t actually shown any business value yet and there’s still a lot of scepticism.
Some people expect the value to be there immediately. I’m old enough to remember the early days of the internet exploding, talking to people who still felt that fax machines weren’t going away. There were people at the time who said, ‘Who’d ever want to do their banking on a computer?’ or ‘Why would you ever do banking on a mobile phone? It’s just ridiculous.’
What’s exciting to me is that the leaders across business and technology that I speak to, the vast majority of them are saying that this really is a game-changer. It’s not just fashionable.
Of course, no one in financial services has yet applied a scaled solution; we haven’t identified one. But we’re going to spend the next year working out where the big bets are.
Mills (nib): Like a lot of these things, too, some of it has been consumerised. My 10 year old daughter’s coming home from school talking about the fact that they’re using ChatGPT in the classroom. This consumerisation of capability through what’s happened with the explosion of ChatGPT has really driven the conversation, and it’s put in the hands of everyone – it’s not just an enterprise-grade tool.
Kaplan (Beforepay): And we still don’t know yet how most businesses will adopt it.
Five to 10 years ago, blockchain was all the buzz. And you had decentralised blockchains and certain conglomerates that tried to own the blockchain; it went nowhere. But similar to AI, you have these open large language models, your Open AIs, that you could put data into and own. They have Facebook [offering] a Llama 2-type model where you can have it on-prem and control it.
What best practice is hasn’t really been worked out, because it’s still being discovered. And then to power a lot of those models or to get value out of them, a lot of larger organisations still need to centralise their data in a way that it can be used meaningfully.
If you have a team of data scientists but you don’t have the data engineers to pipeline the data in, the data scientists can’t really do that much. A lot of that is still being worked out. ◼
This is an edited extract from the Future of Financial Services, Sydney 2023 conference.
