Toby Winchester

 Welcome, everybody, to the next episode of Voices of Sample Management.

Today, we were thinking about having a topic on machine learning and artificial intelligence. Always a very hot topic, in the media, but, how does that affect sample management and even a software company such as Mosaic and their software development, moving forward.

To help me with that, I've asked Gergely Bandi to join me. Gergely, could you introduce yourself?

Gergely Bandi

Hi, everyone. I'm Gergely Bandi. I'm originally from Hungary, and, I'm the head of software development of the UK team for Titian.

I have a… Before joining Titian, I've done a PhD in, virtual living organisms. So that was actually, quite, related to this topic.

And, before that, I've done other software development and at Titian, I started out being a software engineer and then went from there up to now becoming the head of software development for the past 5 or 7 years.

Toby Winchester

And enjoyed every moment of it, no doubt!

Gergely Bandi

Yeah, it was great. It was great.

Toby Winchester

Okay. So, I suppose we should really start off very basics. There's the terminology: machine learning and artificial intelligence, and it seems to encompass, you know, whether you've been watching Terminator films or all the way down to, you know, speaking to your telephone, could you sort of summarise what the terms mean?

Gergely Bandi

Yeah, yeah. So, let's start with the broader one, the, artificial intelligence, which is usually very loosely defined. There are many definitions for it. The one I actually like the best is based on the task at hand, so that that goes like, artificial intelligence is where you apply software solution to a problem that doesn't have a satisfactory software solution available until now.

Which is great because we're moving ahead.

So, last year's artificial intelligence is today's sort of minimum requirement for any phone that you have in your hand. So, you know, we have artificial intelligence in playing chess on an average level, you know, that's no longer artificial intelligence, that's a game.

We had optical character recognition. We had voice transcription. We had barcode scanning or all of these things that used to be AI when they were cutting edge. And now they're just sort of everyday software.

So, we're moving on and AI is the next thing that we are working on.

So, it's everywhere because, you know, whatever you're doing on your phone in your lab, you used to be AI, at one point. So, you know, the way the liquid handler heads move and calculate the optimal way of doing their work.

The way the tube position verifiers use image recognition to verify the positions. That's all AI, or used to be AI.

And then we're making them even better, with new AI, practically.

Machine learning is sort of a subset where you use data sets, past data sets, to do classifications, predictions.

So, it's practically learning from large data to be able to sort of predict the future or the outcome of new data that it's getting. So that's what machine learning is.

It has many types available. And again, it's something that's used in many places. I think one of the big successes of machine learning is image recognition, voice recognition.

These were very early versions of it and we managed to make it work well, sort of a decade ago. So, it's good stuff. And again, it's one of these sort of main pillars of AI as we have it today.

Toby Winchester

And, you see that in rack scanners and you know, even our newly released

Gergely Bandi

Exactly.

Toby Winchester

voice weighing application.

Gergely Bandi

Exactly.

Toby Winchester

Yeah. You can use these technologies because they now have been developed.

Gergely Bandi

Yes.

Toby Winchester

Right. Okay. So, I suppose moving on, you know, how does, machine, yeah… How can you see maybe machine learning being used at, you know, in a software application such as Mosaic?

Gergely Bandi

So, again, it was niche at one point, then we had a few solutions for sort of typical problems, that were used in those types of software. And now, we're at the point where machine learning is so widely available and in so easy ways that wherever you have access to past data and you're really interested in something being classified or predicted, then you can use machine learning there.

So, it's really, find a spot where you could use that sort of outcome prediction and just put machine learning there. So you can use it in small to large areas, so, you can predict, how long an order fulfillment will take. You can predict it at a machine level.

You can predict maintenance, so, that's one thing that they're using now is predict when maintenance will be needed, based on, you know, past breakdowns and such.

So, you can sort of see that you're safer this week than you are going to be next, based on past data. And it does work. So, yes, it's everywhere.

This one is everywhere now.

Toby Winchester

Okay. So, I suppose the, I've heard that machine learning needs good clean data in order to make good predictions.

Gergely Bandi

Yes, yes. So it helps if you have large amounts of data to train it on, otherwise you can do, sort of supervised training where you're just saying this is right, this is wrong, this is dog, this is cat, but even in that case, the more data you have, the more useful it will become.

And in that case, it's a bit harder to get to the point because you have to put in all the effort of training it.

While you can do a unsupervised version where you just have a large data set of, you know, this is how long these orders took based on what about every information.

This is how often these machines broke down based on their type. And then it can just calculate it based on that.

Toby Winchester

Okay. And I'm guessing Titian as a software company is looking into proof of concept on that.

Gergely Bandi

Yes. Exactly, we’re doing sort of multiple, we have multiple ideas of how we can best utilise this and we're working on this at the moment.

Toby Winchester

Okay, cool. So, you're head of UK software development, how does all this technology help software development moving forward?

Gergely Bandi

So, there are multiple ways, especially, generative AI is used in software development.

It can generate sort of the plumbing, the fixtures for your code. So, you write a code. Writing code is very hard. Writing good code is incredibly hard. And it's supported by lots of automated tests. It's supported by lots of infrastructure to make sure it works well and reliable.

And AI is very good at writing these tests, the documentation. So, these parts of the code. AI can now also write, generative AI can now also write code, and people do use that with sort of mixed level of success. So, while it can write a lot of code, it cannot write a lot of good code.

So, all the data that we have today shows that, all the code that AI writes is far more likely to be rewritten in the next two years than those that human’s write. So, you know, it means that it can help, it can give you a boost, but it will worsen the quality of your products.

Hence, use it wisely and use it at the right places.

Toby Winchester

Okay, so, I mean can it be used for the foundations? But then a human developer needs to take over or is the advice at the moment, tread with care.

Gergely Bandi

Yeah. Well yeah. Start with care and always sort of review what it does so it can create, lots of automated test scenarios, test cases for your code. That’s what it’s very good at. And then you can review and add on as needed. So, these type of things are already useful and used throughout.

Toby Winchester

But it doesn't sound like yet it's going to replace every software developer in the world.

Gergely Bandi

No no no, Not in the near future.

So, there are places where AI will, and there's been, very good research into that. So, there's been MIT research earlier this year, that especially looked at, image recognition based tasks and how AI could replace humans or augment humans, in a cost effective way.

I guess cost effectiveness is very important, at least at this time. And the thing they found is that AI can help with, I think around 25% of visual based jobs, but only 3% of visual based tasks in a cost efficient manner. And that's because, it's not easy to install those sort of, you know, image processing instruments at the right place at the right time.

It's not enough to be 80% confident in the results. And these will all increase in the future. So, they predict that by the end of the decade, AI will be able to do 40% of the visual based tasks. And that's sort of what I see across the board.

So, it will always help developers. It will help developers do the tasks that just takes time and not much sort of mental effort. And that's where it's good. So, you can concentrate more on getting the right architecture, getting the right design to make it futureproof.

While AI will do the, sort of, nitty gritty details of the coding that is just practically using your fingers afterwards.

Toby Winchester

Right. Okay. So, I mean, in some ways it's, you know, equivalent to classic sample management started to automate your lab. It's, you know, starting to automate the software development. But it doesn't replace the people. It allows the people to do more useful work.

Gergely Bandi

Exactly, exactly. So, you're not sort of spending your time moving plates between machines. You know, you're actually sort of designing the workflow or you're doing that and making sure that that you get the right result with the right quality that's needed.

Toby Winchester

Okay. So what do you think's the exciting new thing that's going to be coming through?

Will we all hear about ChatGPT and the latest thing from Google, which has a few issues at the moment, I understand, mainly because they're trying to put rules on top of something that is supposed to be rule-less.

Which, is an interesting problem. But you know, what's new and fascinating coming in the future?

Gergely Bandi

Well, I guess it will more be what the effect is on our lives, rather than what will come.

So, as everyone says, I'm quite afraid of sort of, you know, fake news type things, which is now incredibly easy to do. Some of these systems have built in safeguards against it, but there are lots of local, solutions for this.

So, on my laptop, I can generate images, videos or text with my own local AI here, and I make the rules for them. So, you know, the fact that you cannot, maybe cannot use ChatGPT to create sort of false news with which false images doesn't mean that you cannot do the same locally. So, if someone wants to do that they can, they, you know, have the tools.

So, I suspect that will affect our lives greatly.

Toby Winchester

But could you have an artificial intelligence system to look out for artificial intelligence created?

Gergely Bandi

Yes, yes, yes, it is an active, part of research at the moment and there are tools to sort of spot that. I suspect they will be harder and harder to do so.

And I guess it will go into the impossible level, though, yeah, it will be a threat on that front, but I suspect much less so in our jobs.

So, as we talked about, I think it will augment what we do. It will make us be more productive because we can do more of what matters and that it can sort of do the rest, which probably means that people will need to be slightly more trained, specialized to, you know, use that as the way that actually gets the value out of people, because they will be able to do more.

So, you know, in terms of software development, that means that we can write more software with the same amount of people so we can put more features out. It doesn't mean that, you know, it's going to take less people to do what we're doing. It just means that we will be able to do more for our customers than before.

Toby Winchester

So the pipeline becomes quicker rather than…

Gergely Bandi

Exactly. Yeah.

Yes, So there's not a finite amount of work. That's something that sometimes crosses people's minds. There's no finite amount of work. There's just finite amount of output and effort you can put into it.

So, you know, that's what the Industrial Revolution showed us, that, you know, all the things that got automated didn't mean that people cannot really do work, is just that they can do more and produce more, with the same amount of work.

So, I don't think we need to fear about that, but and again, in the lab, people already know because a lot of things have been automated, a lot of things Mosaic has automated and improved, and it just means that they can actually do more work, rather than, you know, needing fewer people.

Toby Winchester

Okay. Cool. Thank you. I haven't got anything else to add. I think that's a good little chat. Thank you very much, Gergely.

Gergely Bandi

Right, happy to help.

Toby Winchester

Cheers, bye bye.

Gergely Bandi

Bye

Toby Winchester

And thank you, everybody, for listening in.