The Promise of AI with Julia Ferraioli at AIE 2024

Welcome everyone. This is the kickoff to the AI development track. My name is Julia Oli.

My pronouns are she her, and I am an open source AI ml strategist for AWS. My background is actually in ai. Um, back in the day, uh, many moons ago, I did research into novel ways to explore high dimensionality data using artificial intelligence techniques.

And this was all done with open source software. So it's a nice way to come back around to my beginnings in talking about artificial intelligence and where we see it going. So this is a promise, the promise of AI talk.

We're gonna look forward into what we see today and what we can expect in the future. But you can't know where you're going without at least having a little bit of idea of where we have been. So let's take a brief look back and we are going to go back all the way to early history to the pre 17 hundreds.

Mythology and fiction were rife with automata and artificial thinkers. There was this idea that statues were capable of human thought in ancient Egypt imbued with power from the gods. Deads is known for creating statues that were able to move and operate on their own.

And there are many more examples. Jumping forward to the 18 hundreds, philosophers are starting to think that maybe we can codify rational thought in a say in the same, in the same way that we are formalizing mathematics that really paves the way for artificial intelligence there Forward. In the 1930s, biology saw breakthroughs in the neuroscience field specifically about how neurons work and the idea that these neurons which operated a lot like machines, might be able to be simulated.

In the 1950s, Alan Turing published his seminal paper that proposed the Turing test, which is when you're talking to a machine, can it fool you into thinking you're talking to a person. Now, we passed the Turing test not terribly long ago, and new tests have been proposed. And it wasn't until 1956 that artificial intelligence was established as a field of study at a Dartmouth college workshop.

Before then, it was still academic and actually it was still academic for some time after that too. But this is when it became a thing. The 1960s saw massive advancements in artificial neural networks, natural language processing and robotics.

There was a lot of funding allocated to that research, but it dried up in the 1970s with the first AI winter. And this is in part because the expectations for that research didn't live up to reality. It still, it still stayed firmly in the domain of research.

It wasn't, um, productionized. The 1980s saw a comeback with expert systems and chess, which had been held up as a benchmark for ai. Given its complexity in search space, artificial intelligent, artificially intelligent chess players won matches against a chess master, but that wasn't enough.

And we hit the second AI winter. And this was due for, due to a, a variety of reasons, including, um, just the economic climate at the time in a massive oversimplification. We now have the introduction of the cloud and the cloud really changes what we think of as artificial intelligence.

And why is that? Well, it shifted the development paradigms. It changed how we developed what we developed.

We found new challenges, new opportunities, and importantly, there were three factors that made this advent of cloud a game changer for ai. These three technological shifts have led us to this moment. First, the advent of inexpensive computing with a pay-as-You-go go model similarly low cost storage with a pay-as-you-go model.

You ne you didn't have to have data centers full of machines full of storage that you maintained perpetually in order to do research. Both of these service revolutions were coupled with hardware advances. Our applications scaled and we were collecting more data than ever before.

And storing it for future analysis became low cost, almost marginal. This changed things for modern AI because prior to this, our challenges were access to computing and availability of relevant data. Do I need to go build a Beowulf cluster or schedule time on my institution's supercomputer?

How do I parallelize my machine learning in a way that works for the infrastructure that I have? Those questions became obsolete with the advent of cloud. What did this mean?

Well, we saw a increase in innovation and a rapid acceleration of the pace that we could make advancements research and artificial neural networks leapt forward. We could now construct massive networks with layers upon layers upon layers of neurons. And those networks were able to succeed at intractable problems or previously intractable problems including strategy.

Game. Go Go is a um, a board game that has a much larger search space than chess and AlphaGo was able to beat professional go players for the first time. Deep learning could uh, be applied to data that had been previously thought to be too computationally expensive like images and videos.

It could understand them, create them and modify them. And for the first time we had really good image and video recognition models that achieved unprecedented accuracy levels, including classic benchmarks that had been, you know, that they had been working okay against those benchmarks. But with these deep learning techniques, they achieved near human performance.

And it wasn't limited to research, it was shipped to consumers, putting hand, putting machine learning capabilities in the hands of anyone who can access an API. This moved foundational machine learning functionality from research to production and it was applied to domains far and wide, including knowledge-based systems. And the best example there is when IBM's Watson defeated Jeopardy champion Ken Jennings, which is no small feat from a, from a technical level as well as Ken knows a lot.

So this ushered in what some might consider a golden age of software development. We have more tools than ever, more challenges than ever, but this is a promise of AI talk. So we're gonna focus on the opportunities we can say out with the toil and in with the fun.

AI is helping us spend our time where it's the most valuable. We can give it boring but important tasks like writing tests and put ourselves in the shoes of a code reviewer and adding in our expertise where needed or have AI help us develop models for the systems that we build and check for fundamental design errors. We can ask it for information that takes into account our context and environment so that we're not overwhelmed or distracted with irrelevant data that makes our days longer.

It helps us learn and develop agents, have a better understanding of where we need to grow. We can teach more people to script and code and we no longer have to consume entire manuals or documentation sets to get the answers that we need when we need them. It can synthesize massive quantities of information and provide it to us with references, using techniques like rag with AI assistance to answer questions or lend a hand when you need it.

We have more time to specialize, create, or actually take the vacation that we've been meaning to take for quite some time. We can think of these capabilities like there are new calculator, they're designed to make life easier. They're not designed to replace knowledge or or expertise, but simply facilitate access to functionality like standard.

The standard libraries in various programming languages, co-generation abilities included in open source frameworks help us with syntactic understanding like the ones that we find in ides and plugins and simplify development. With tooling integration, we're really still just getting started and AI has a long way to go and a lot of potential, a lot of promise. So how might it change going forward?

It has so much potential for us as software developers. It can help us understand and navigate large code bases which seem to be getting larger and more complex. With every passing hour and every passing commit, we can potentially automate the modernization and migration of dependencies, languages and frameworks.

It can help us learn new concepts or branch out into new areas, figuring out new challenges for us and preventing us from stagnating if we can quickly spin up new ideas, test them out, and find the optimal solution with less cost both to us, our companies and our users. But it's not just how it changes things for us as software developers, it's also how it changes what we create and how we create. I'm looking forward to increased access for more people with machine translation getting ever better.

Increased access for all populations with accessibility gaps being highlighted at every stage of the development cycle and helping us improve our core interpersonal skills for better collaboration and better team dynamics. Really when it comes to the promise of ai, the field is wide open in this track. You'll get to hear about innovative approaches, capabilities, and applications of ai from fine tuning techniques to prompt engineering to ML ops and race cars.

Even a great program awaits and I hope you enjoy. Thank you so much. Um, you can find me at all of these various links and I hope to see you at a conference in the future as well.