I’ll be looking at generative models and how the public reacts to them. I remember back in 2019 when we encountered the first GPT models. It was impressive in what it could generate but not very useful. Fast forward today - generative language models like InstructGPT (GPT-3.5) and its sibling ChatGPT are finding their way into all sorts of applications due to their capability of generating reasonably sounding and fairly accurate outputs from the user’s instruction. The decision for OpenAI to include the public in ChatGPT with the easy-to-use interface has suddenly made a lot of people who aren’t in the AI space realise how far things have come since the first chatbots of 2017 (when we first started in chatbots).
They’re also reaching a point where they’re good enough for people to be concerned, and for good reasons. The outputs are becoming so convincing that it’s really hard to differentiate whether it was generated by AI or a human. This can create problems for sites like Stack Overflow (who recently banned ChatGPT answers) because it can create legitimate sounding answers that may not necessarily be correct. And due to the speed at which it can be generated, their community review team can’t keep up.
Regulation may also come in to manage the proliferation of AI generated content, due to the potential harm it can bring.
Something really cool is that over the last few years the use of AI has become more accessible to nontechnical people through the development of no-code, drag-and-drop, and prompt-based engineering tools. This means that anyone, regardless of their technical background, can create solutions using AI. In the next few years, it’s likely that we will see even more tools and platforms emerging that allow nontechnical people to build with AI without having to study for four years. This is really cool because it opens up the potential for more people to take advantage of the benefits of AI and create innovative solutions.
I’ll be watching several aspects closely. For one, I am personally concerned about how it is finding its way into the defence sphere and how current conflicts around the globe are seeing the introduction and testing of AI based weaponry. So I will be trying to keep an eye on the publicly available information about this.
Closer to home, my interests are more commercial. I am particularly interested in new commercial deployments of different types of Artificial Intelligence here in New Zealand. There are so many possible use cases, but limited capital - human and financial available to deploy.
There is a lot of power in foundation models. These are large AI models trained on a giant amount of data (often internet scale) which have generalized enough knowledge to be used for various downstream tasks. GPT-3 is a perfect example - the same model can be used for traditional NLP tasks such as sentiment analysis, text classification, intent matching, etc and often better than traditional models due to the vast quantity of knowledge it has ingested during training time. Stable Diffusion (images), Whisper (speech to text) are also examples of foundation models. I think that businesses will begin to leverage these models in situations where they would never have enough data to train a quality model, particularly in the contact center where there is lots of power in parsing customer queries but the data available to train a model is of poor quality (skewed, transcription errors, etc). It will allow them to realize the value of AI much quicker than before. There is also much less capital expenditure required to leverage a foundation model.
One potential use for chat-based AI systems like GPT-3 is as an intermediary step between fully automated customer service and human responses. While many businesses may be hesitant to completely replace their customer service teams with AI, chat-based systems could be a useful tool for keeping customers engaged and entertained while they wait for a human representative to become available. Businesses could also use chat-based AI as a way to brainstorm and generate new ideas, asking it questions and using its responses to spark creativity and come up with new solutions.
It seems to me that some of these new large model releases are opening the door to small business innovators to experiment with the technology. It has previously been noted that there is a risk that significant commercial benefits from AI will accrue mostly to larger organisations who can afford to design and build complicated models. Just last week I saw an example where a small business owner had experimented with ChatGPT to build out several AI generated tools for marketing. This was a great example of a move towards democratization of access to the benefits. I hope to see more of the same opening up in 2023.
I’m particularly excited about the proliferation of AI in products and services. People have been talking about the democratization of AI, and I think foundation models, particularly open source versions, put the power in startups and smaller organizations' hands. What was too costly and time consuming to develop, and probably even infeasible in terms of accuracy 2 years ago should hopefully be unlocked by foundation models. Seeing a cleaning business leverage ChatGPT to speed up their operations is just the start of it. There is also an increasing number of companies and organizations looking to own the generative space - the generative AI arms race has begun and is going to continue in 2023.
Getting our own office space again. One of the best parts of the last year has been the opportunity to do cool events together and to just hang out and have fun as a team, so I’m looking forward to being able to do that more often in 2023.
My favourite would be an end to Covid restrictions, but there will likely always be a need to protect the vulnerable section of our community - so that might be more of a wish.
On the AI frontier I am looking forward to better customer experiences delivered at scale. As a customer of many large service organisations, (including Government services) I personally shy away from some tasks simply because I know the experience is going to be an unhappy one. (like having to listen for ages to hold music so loud it is horribly distorted) so I hope to see within 2023 some of those experiences made faster and more enjoyable.
Also on the AI front I look forward to seeing more boards and execs engaging more deeply in understanding the technology and how it might impact their business models. Our offshore competitors are investing and NZ has to move fast to ensure not being left behind.