The Future of Applied AI

... and how to make it work for you!

Artificial Intelligence (AI) seems to be in the news all the time. The rise of impressive Large Language Models (LLM) such as Chat GPT and the wide range of realistic generative artwork models have stoked the excitement and fears that AI typically brings. But where is it going? And what does it mean for you?

The majority of the advances in applied AI are in vision and text understanding, ie the data is an image or document rather than a row in a database table or Excel file. This is called non-tabular data and needs deep learning models running on large speciliast chips. The most sophisticated models run billions of parameters in their models across huge collections of documents and images. The models have complex design and hyper-parameter tunings, run across thousands of servers and run over and over again to get improve their results. This needs a vast amount of data, compute power, electricity and spend. This is why these models are built and managed by large / mega tech with huge budgets.

The success of these models means that AI is bifurcating. One branch of AI is driven by huge budget, specialists that experiment with wide application "general understanding" models. The other branch is applied AI - understanding the best approach and how to design a project that will deliver value for a particular "use case". The two branches cross-over when general use models are adapted to do specific activities (and achieve even better results). This has led to the rise of what is now known as "prompt engineering".

So how do you get the most out of AI?

1. Existing software - the good news is that it is helping you already. The software that you use to write a document, create a spreadsheet, presentation, host a conference call or plan a project is already powered by tools driven by AI. All the auto-filling, suggestions, recommendations and "co-piloting" are AI powered. They will continue to evolve and get better.
2. Using new AI tools. Querying large language models can be used to generate new ideas and content, check grammar, spelling, meaning, translate or synthesize something that has already been written. New AI tools can dramatically increase your productivity when writing, reading, coding or looking for new ideas. To get the most of out, you need to rethink your ways of working and consider how much time you may be able to save by using LLM's.
3. Building bespoke AI. General understanding models are good, but bespoke is nearly always better and faster. Organizations have "data moats", or data they only have access to. They also have specific and often unique processes that work better when AI is trained specifically for their use. There are numerous techniques for getting the best from general models and custom models. This is where applied AI design really comes into it's own.

Most areas of AI continue to thrive. However some areas are losing relevance due to the rise of the image and document understanding models. These include sentiment models, term frequency models and simple object classification models.

So where's the value in applied AI in the future? The recent development of AI has helped fast track development and underlined the importance of key factors in applied AI:

• AI solution design and architecture
• Business and process understanding
• Agile, iterative delivery
• Model updates, learning loops and "human in the middle" steps and quality assurance.

And what about you....

Do you still have a job in an AI future? Yes! An even more exciting role that removes tedious tasks and helps you build on your ideas.

Should you hire even more data scientists? No! Advances by large tech mean you should be able to evolve your data science spend towards bespoke applied AI. This is how you will keep your competitive advantage and get value from applied AI.

Algospark is an expert in applied AI. Get in touch to discuss how we help you get the most from data, maths and technology.