The Power of AI vs. the Power of Trust, Models, Architecture, and the App
VOLUME 1 - ISSUE 9 ~ AUGUST 2, 2023
The JC2 Ventures team (John J. Chambers, Shannon Pina, John T. Chambers, me, and Pankaj Patel)
We are in the thick of summer – it’s hot, and the AI revolution is even hotter. I am excited to share the next edition of the “CIO Two Cents” newsletter with you all. Read on for insights from me, Yvette Kanouff, Partner at JC2 Ventures, into what’s on the mind of CIOs at this moment in time.
As we all know, AI has been the biggest trending tech topic of 2023. While the world is getting up to speed on the “wow-factor” of tools like, OpenAI’s ChatGPT, Google’s Bard, and Meta’s LLaMA, I’d ask you to think beyond the hype. Companies need to consider now what the AI wave means with regard to trust, ethics, performance, and data sovereignty/privacy.
AI is wonderful, and I strongly believe it will transform the way we live, work, and play. That said, AI is not new. I was writing AI algorithms in the last century, but what excites me today are the ‘tools’ that have recently been created and the power they have to transform enterprises. I can now call an AI API and build applications faster with AI tools like ChatGPT and others. Let’s examine the impact this will have for businesses a bit further.
AI Models vs. AI Applications
It’s common today to find companies that use AI. This means that they likely use an AI model, such as GPT-3, Bloom, or others. These models are trained using significant data, and they are readily available for companies to use. Some companies use custom or self-built models, while many other companies use multiple models to triangulate on the problem at hand in the best way. The key question, today, is not if you use AI, but ‘how’ do you use it, what and how many models you use, and how this gives the most accurate results. One differentiator in models is specific training for a use-case, which implies less erroneous outcomes. Keep in mind, however, that building and running AI models is costly.
Scale, Architecture and Cost Structure
On the topic of cost, one point to emphasize is a company’s architecture and scale. We hear daily of the compute intensity of AI, cloud architectures for scale, and more. AI is a ‘gas guzzler’ when it comes to cost. When looking at AI solutions, it’s worth double-clicking on the architecture to ensure that it takes compute costs, speed/latency, and scale into account.
Trust, Ethics, Data Sovereignty, and Data Privacy
There is much discussion worldwide about the ethics of AI. If a model is used that includes copyrighted or unethical information, it may result in the same – and there are regulatory requirements to consider. Even if they are not in place today, they will be soon, and models should be built with this in mind. Today’s AI choices should include legal discussions and the topic of trust.
For example, how will my data be used? Will it be used to retrain models to use with other companies? Will my data be protected in storage, transport, and use? Are there ethical or legal issues in the model(s) used that might cause concern in the future? Trust, compliance, and data quality, sovereignty, and privacy will be major factors in AI solutions in the future, especially as regulations catch up to the speed of innovation.
Usefulness, Ease and Differentiation
Given that AI is everywhere, we also cannot lose focus on the outcomes of the application using it. The goal is to create simplicity, ease, and differentiation in solving problems. I am now able to find network, customer, medical issues, and more, that I have never been able to find with human analysis. AI is powerful and amazing, but the implementation must provide an outcome that is differentiated and simple to implement. For the consumer, Alexa, Siri, DuoLingo, FitnessAI, Waze, otter.ai, and many more tools are already changing the way we go about our daily lives. And there is much more to come, that will leverage, in particular, how enterprises utilize and adopt AI into workflows.
There are so many tools that exist to make the use of AI easy. Building an AI product can be done with low-code, no-code solutions. Transformer and RNN (recurrent neutral network) models can be downloaded and used quickly with amazing results. That said, it’s worth looking at the training sets, legal and security implications, architecture, and differentiation. We will continue this journey. I don’t think there is any turning back – the future will be focused on trust and outcomes.
Well, that’s my 2 cents on AI as it stands now – and there’s certainly more discussions to be had as we continue to watch it evolve in the future.
Moving fast? I've got you covered:
Today, the key question about AI is not if you use it, but how you use it, which models you use and how many, and what will give the most accurate results.
When looking at AI solutions, it's worth double-clicking on the architecture to ensure that it factors in costs, speed/latency, and scale.
AI is everywhere. We not lose focus on the outcomes of the application using it.
