With an expected annual growth rate of 37.3% between 2023 and 2030, AI will undoubtedly revolutionize every industry and profoundly impact the way we live, work, and interact. With this transformative technology, businesses are unlocking new levels of efficiency, innovation, and customer satisfaction.
With this as a backdrop, Tech in Motion brought together recognized experts (Moderator Justin Grammens Founder & CEO @ Recursive Awesome; Frincy Clement Principal Data Scientist @ ADP; Victoria Vassileva Director, AI Performance & Responsibility @ Arthur and Omar Rahman Sr. Director, Machine Learning & Detection Platform @ Salesforce) to discuss AI’s impact across different tech sectors, concerns around ethical AI, and AI and machine learning opportunities in the job market.
Read on for highlights of the conversation, and to watch the full event, click here to get the recording sent straight to your inbox.
What is the impact that AI is having on your industry?
Frincy: Pioneers are all talking about AI achieving superhuman characteristics. This year I have never seen this kind of interest. Because of the emergence of a simple AI in almost any industry, the shift in thought processes has been amazing. Previously, we had to convince decision-makers to invest in AI, but today, there is a top-down push like we want it tomorrow. That has helped tremendously to drive acceptance.
A lot of curiosity has come from the non-technical audience. Project managers, sales, and marketing individuals want to transition their careers, not as technologists, but as participants in the AI revolution. They are interested in the more practical aspects of AI and the tools they can use.
From an ethical standpoint, the role of responsible AI has stood out particularly, because previously the function existed, but companies today are looking closely at frameworks, policies, governance, compliance, and other business concerns.
This has caused somewhat of a shift in responsibilities. With generative AI being such a hot topic, there has been a huge interest in specific skill sets needed, leading to a skills gap among needed practitioners.
Victoria: There’s no industry left unturned in terms of where AI is being felt and embraced and impacted.
I work across a lot of industries. Generative AI is a hot topic with a number of tools available. It has democratized access and allowed people to play around with different models.
One step behind this is compliance and ethics standards. Governments and companies are forming task forces to keep pace, but they are behind the curve. There is no limit to access. Every company has data of some sort. They are looking at how can they access this and provide it to their customers. How do they secure this data and put guardrails around it? What are the implications? How does it impact jobs? A lot of this dust is unsettled.
Justin: It can almost be perceived as general-purpose technology like electricity which can be plugged into from many places.
Omar: People are saying AI is taking the world by storm now, but it’s actually been happening every few years. If you go back a few decades, there are some inflection points that have happened in the field of AI.
In the cybersecurity industry, we deal with a lot of anomaly situations, to determine if any threat actors are getting into our systems. Credit card companies do it. Gmail does it. Examples of anomaly detection apply to many different situations. We operate on tons of data. We get lots of it from default from analyzing logs.
ML is the right way to approach these problems. The rules are too many to hard code the solutions. We cannot build specific rules around when a person accesses data, and from what computers, so we are using AI to catch more bad actors than in the past.
On the flip side, threat actors are also using these techniques to develop phishing emails and other scams. Now they just use Chat GPT to develop their emails. It has become basically a continuing arms race to achieve supremacy.
How have you seen companies govern generative AI to balance risk and value?
Frincy: When it comes to generative AI, we are at such an early stage right now. Everybody is still trying to understand it. You need to be operationally mature to benefit from generative AI and every kind of AI.
Everybody is interested in benefits and that has been discussed a lot. Differences between traditional AI and generative AI. Even before an organization gets into generative AI, it needs to think about readiness. Need to achieve AI maturity in terms of security of system data.
The most important thing is data readiness. Scalable, retrievable format ready for AI, and processing. Do you have the right resources on your team? I have seen a lot of companies transition to the cloud to gain efficiency, but once they are there should be enough support for employees to understand to think twice about why we are using AI. Who will it affect? What is the impact? How do we monitor performance? These are general considerations.
Read More: How ChatGPT and Other Automation Tools Can Make the Tech World Better
With generative AI, the builders of these should know what guardrails and governance they need. There's a lot of limitations. It can give you answers that are inaccurate, called hallucinations. Face recognition is an example of biased data that the model has been trained on. Cybersecurity threats, make sure private data is protected.
Companies can start playing around with guard rails to make sure private data cannot be accessed. What is the impact on business before putting it into production?
If you have the infrastructure in place, you can start with a base model and train with your own data. Eliminating bias in data and relying on good data in your enterprise. At the same time, we have enterprises creating APIs, so no data is leaked back into other systems. Establish feedback loops.
Once there is confidence in these processes, it can be scaled into multiple applications.
Victoria: Catching more straightforward things like sensitive data, is easier than something like domain-specific areas. Bring in human stakeholders into the applications because not everything will flag for transparency, governance, or bias. As you ask more complex questions, the need to include human intervention is more important.
There is a lot of complexity around models. May want to flag attempts rather than block them to investigate further, inside companies. Human judgement is not clear cut when it comes to opinions. This will be an ongoing iterative process including a company’s brand values, and managing legal risk around each situation. Do not expect these language models to be always right.
Justin: The collaboration between humans and AI is the best way to go. Expect lots of little mistakes.
With Gen AI, how will this affect the software engineering discipline?
Omar: It will empower software developers to do more. Just like using Google to help you write code, except this is much better with large language models and generative AI. No worry right now that developers will become extinct. It will evolve the job as programmers still need to view the results and understand them. These deeper skills will still be needed.
AI has created many roles, especially in data science and data engineering. I don’t think any of them are going away.
How will people keep up to date?
Omar: Be aware of application program languages. Get your hands dirty with programming languages and algorithms. Keep up to date with publications.
Frincy: Evolution is what we need to think about. People most impacted by AI will be those who run away from it. Being part of a community helps in terms of being up-to-date and attracting like-minded people. Network and learn from people. Every other week something is happening. Meet in small groups and listen to summaries of small research that can be presented.
Be curious and play with tools. At the same time, even if a job is partially automated, the creativity that humans can add, cannot easily be replaced by AI. Data, which is trained, you cannot introduce any outliers to it without human thought.
Victoria: Embrace it. Humans with AI will replace humans without AI. The biggest benefit is productivity. What are we doing with the extra time? It is up to the organizations to determine. Start asking questions. There is room for everybody. Explore these tools. What are you trying to accomplish with a model?
Omar: While there has been a huge swing to larger and larger parameters, performance doesn’t need to be driven by larger models. Better data comes with smaller models. When you think of applications, one model to do everything is impractical with different data compliance requirements and individual nuances by industry.
The future is a bunch of finely tuned models within vertical sectors such as health care, finance, manufacturing, and many more.