Key issues affecting the effectiveness of Generative AI

Generative AI looks so exciting. However, it carries significant issues that will likely cause initiatives to fail or substantially underperform against their potential. This blog presents information on various topics. We’ll first address a key issue that causes much resistance to the adoption of genetic AI: the technology presents a probabilistic answer as if it were a deterministic one. This blog will help your company better understand where and how to apply genetic AI.

Companies report that they have faced a lot of resistance and a surprising number of change management issues when trying to implement genetic AI in their business. They can commit senior leadership to it, but middle management and line people resist. Some understandably resist because of the possibility of job loss or the need to fundamentally rethink how they would do their jobs. However, a significant amount of resistance comes because companies struggle with the issue of probabilistic versus deterministic responses when using the AI ​​tool.

AI responses are probabilistic

Generative AI provides a probabilistic answer, meaning it provides the most likely answer or next step. Therefore, when he composes letters or a written work, he provides the most likely next word, phrase or idea. However, the most likely answer is different from the correct answer. Granted, it is often a correct answer, and – with access to massive training data – has a high probability of being a good answer. But that is very different from a correct answer.

Further complicating this issue is the audit trail challenge of verifying how the response was derived. Many of the challenges faced in applying genetic AI come from applications where a deterministic answer and a complete explanation of how the answer was arrived at is required. Where answers are needed, or believed to be needed, as deterministic, organizations and the people in them resist these applications as they cast doubt on the usefulness or reliability of the results.

This proves to be the case even when artificial intelligence is offered as a co-pilot and a human is called upon to be the final arbiter of the answer. Lack of explanation and uncertainty around help creates mistrust and resistance.

Just because genetic AI presents probabilistic answers doesn’t mean it isn’t valuable. There are many use cases where it is valuable. However, if a company applies this answer to a question that needs a deterministic answer, then it appears to be lying.

Companies experience frustration as they try to apply genetic AI to their business because they end up using probabilistic answers when they need deterministic answers. So they need machine learning in conjunction with a probabilistic principle (like, it’s a tree because it has green leaves), and then they need to test to see if there are issues (like autumn) that affect the effectiveness of that answer.

What areas are good for probabilistic AI models?

How should companies introduce genetic AI into programming? The tech industry has made some bold predictions that genetic AI or AI can learn to code and dramatically improve the efficiency or productivity of coders. Well, that’s interesting.

There are areas in coding where this is immediately effective. For example, testing test case development is something that can inherently be handled through a probabilistic model. What is important here is that a company can test for many conditions and create scenarios as wide and deep as possible. Creating effective tests for code and security breaches is essentially a probabilistic exercise. Hence, genetic AI shines and is easily adopted by professionals.

On the other hand, it’s not good for actual code development. A company wants the code to be correct 100% of the time. This calls for a more deterministic answer. However, it’s great at knowledge management that supports code development and can also create great bootstraps that greatly boost productivity.

Another aspect that genetic AI is very good at is knowledge synthesis or summarization. Therefore, areas around knowledge management are ripe for genetic artificial intelligence.

For example, within the CRM function, Salesforce is making great progress with Einstein products and other products aimed at synthesizing or summarizing information from customer requirements information. Making salespeople more efficient with knowledge management is a very effective use case for genetic AI.

Marketing is another area where a probabilistic answer is great. Example: What is the likely best next step to take with this customer? Genetic AI can have a huge immediate impact when sorting through customer data to determine the best solution to customer problems. It can be a powerful tool there.

In contrast, a deterministic response is necessary in processes such as claims processing. Example: How do we get the right amount of money for this health care claim? This requires a deterministic answer. Getting it right most of the time is not good enough. it must be correct all the time.

An effective start to using genetic AI and reducing frustration and resistance is to ask the following questions:

· Where can we use it productively immediately?

· Where does it need to marry with other technology?

· Where should we not use it?

What about the human equation?

While a useful tool for problem solving, a probabilistic answer is a place to start. However, it often won’t take people out of the equation. Instead, it equips people with much more sophisticated tools, particularly when a company feels it needs to arrive at deterministic answers.

A company can reduce resistance if it uses the AI ​​tool to help organize data and summarize data. This is helpful. But when using it to make decisions, people can feel uncomfortable with the decisions because they are not always right. And since they’re not sure how the AI ​​tool got its answer, they also don’t know how to check it.

What is the solution to these dilemmas?

The solution is to gain a fuller understanding of not only the work being automated or assisted, but also how it affects other human work and provides guidance and assistance with downstream work or other unintended consequences.

This requires becoming more mature in where the tool will be applied. Users should exercise caution and consider the implications. In addition to assessing the maturity of the product, it is necessary to consider the people and the organization and the unintended consequences so that they do not resist and kill it.

Some issues that require this thought process include:

· Consider not only the automation of work but also the implications of a person’s role being automated.

· Consider the reality of any productivity improvements required that will result in fewer people doing the work. How will you allow it?

· Consider how to test the tool so people can trust it. The more uncertainty a disruptive new technology brings, the less people want to trust it.

· Consider how to pilot the AI ​​tool to demonstrate its benefits before moving to full-scale changes.

Some Final Thoughts

After a year of frantic experimentation with genetic artificial intelligence, the industry has successfully conducted thousands of pilots. However, most of these pilots failed to go into production because they were hampered by the myriad challenges of funding, change management and adaptation.

However, where genetic AI is appropriate, it moves quickly into production and delivers impressive returns. The success of these production use cases appears to be driven more by the fit of the use case to the nature of generative AI than by choice of tool or other factors.

Furthermore, it appears that for most business functions, there is a productive role for productive AI, but focused on the right sub-function. Given the huge investment made for experimentation, perhaps the most effective way for most companies to discover where genetic AI will be effective is to look at use cases where genetic AI has successfully transitioned from pilot in the production. Where this development has been done en masse, companies can be confident of good implementation and substantially reduce the risk of wasted effort and capital.