Soaking in the Pressure of Using AI

When large organizations spend billions of dollars in research and development of a revolutionary technology, a time comes when the technology is ready for prime time. The technology giants put their best foot forward to ensure large-scale global adoption of the technology. These are exciting times for any technology enthusiast, and it is natural to feel the urge to be a part of the bandwagon and not feel left out.

People across the rank and file of an organization are feeling the pressure of using AI-based solutions for every business problem. How do you soak in the pressure and make the right decisions for your business problems? Let's break down the problem and remove the cobwebs to get a clear picture of when to use AI and when not to.

Fundamental Questions to Answer

Before you get too deep into the weeds of designing solutions for specific business problems and use cases, there are some very fundamental questions you can ask yourself that will give you very strong hints on whether you need to invest time and energy in building AI solutions.

• Will investing in an AI-based solution help us disrupt the status quo of our industry?
• Will it help us leapfrog from an industry laggard to industry leader for any of our products or product features?
• Can we use AI for any of our product development or product operations processes? Does it significantly reduce the cost of operations in the long run for our product?

While this kind of questions are often brought up and answered in board rooms at the executive level, in good companies with a strong product culture, decision-making is democratized, and individual teams are empowered to make key decisions for their product roadmap and product features. 

Hence, it is important for technical product managers and their teams to think like a CEO and ponder on these fundamental questions when deciding on using AI-based solutions for problems provided by customers and executives.

Peeling the Layers

Now, let's take the process of questioning and analyzing a layer beneath the fundamental questions and use a framework that will help you decide whether to use AI in your solution design.

Predictive AI

While there is a lot of content in the public domain on the difference between predictive AI and generative AI, I still want to harp upon the difference between the two. Generative AI (GenAI) has garnered so much attention in the last few months that people often overlook the power of other types of AI, and investment and innovation in those technologies have slowed down in a lot of organizations.

It is important for development teams to pause and think about creative solutions for the user/ customer problem at hand. 

Is the problem you are trying to solve related to gathering historical data of past events, finding patterns in it so that you can predict the outcome of future events? For this kind of problems, predictive AI is where your solutioning river stream should branch out. 

If you are thinking about Gen AI for such problems, you are barking up the wrong tree. Some common examples of problems that can be solved using predictive AI solutions include financial forecasts, optimum infrastructure utilization, fraud detection, etc.

Once you have clarity that the problem you are solving will require a predictive AI-based solution, what kind of machine learning algorithm will solve your problem should be your next logical step in designing the solution. 

• Classification. Assigning data points to predefined classes is a classification problem. For example, flagging any content on a social media platform as inappropriate so that the machine knowledge of appropriate vs inappropriate content is continuously refined is a classification problem.
• Regression. When the goal is to define the correlation between features and your target variable to predict the future problems is a regression problem. For example, predicting the future traffic on your service to decide when to increase the number of pods.
• Clustering. Grouping data into different buckets with the goal of defining the right buckets that will help you make decisions is a clustering problem. For example, analyzing customer behavior and use cases to help your organization define the right customer segments is a clustering problem.

Isn't it quite amazing that there is so much innovation and creativity you can drive within the realm of predictive AI? So far, in this article, we haven't entered into the realm of generative AI, which has created so much buzz lately.

Generative AI

In contrast, if the problem you are trying to solve involves training an AI model on your organization's raw data, making it learn by example so that it can take prompts from a user and generate novel outputs, that is where GenAI comes into the picture. The problem might have different permutations and combinations of user prompts and the generated output. The user prompts could be speech, text, or unstructured data, and the generated output could be natural language text, speech, images, or videos.

Common examples of problems that can be solved using GenAI include customer support, code generation to reduce product development costs,  data synthesis for research and testing, creating on-demand marketing content, etc.

When it comes to GenAI-powered solutions, once you are clear that your problem requires a GenAI-based solution, there will be additional important solution design decisions that will help you build the most impactful solution.

• What LLM base model should I use?
• What are the latest architectural frameworks and design patterns that i should take advantage of. Think RAG, RLHF, multi-agent react agent using langflow, etc.?
• How do i ensure the right guardrails of security and data privacy?

Conclusion

While a lack of technology modernization can make a company's business model obsolete at some point, it is important to ensure that you are using your organization's AI investments for the right problems and at the right places. Forcing AI-based solutions for problems that don't need them can take your product in the wrong direction. Keep your calm and use AI investments wisely.