How can you encourage AI experimentation in your company?

The goal isn't to seek out problems where we can fit AI solutions but to identify existing challenges where AI can be the most effective remedy. Our strategic approach should focus on understanding the capabilities and limitations of AI technologies while fostering an environment open to experimentation and testing. Initially, our focus should be on areas where AI can enhance our efficiency—particularly in repetitive tasks—and those where it can improve scalability. It's crucial to remain mindful of the technology's current limitations, including concerns about data accuracy, privacy, and associated costs. We aim to kickstart our AI journey by tackling small but high-impact problems, benefiting our operations and customers.

We begin with the importance of having a clear vision and aligning it with your growth strategy. Education is key, so we explore how to bridge the knowledge gap within your team. Embracing a culture of innovation and accepting failure as a stepping stone to success is also crucial. We touch on the value of having a dedicated AI team and securing your data fortress. Collaboration with startups and pilot projects is a must before diving into full-scale AI implementation. Ethical considerations are highlighted, and we wrap up by emphasizing the dynamic nature of AI. The bottom line? AI can work wonders for your business, no matter its size. So, don't wait—dive into the world of AI experimentation and transform your business!

When developing AI products, it's misguided to force-fit problems into the technology—a longstanding error. It's essential first to pinpoint industry challenges or product needs, then find the best technology for solutions. While many issues are addressed using machine learning or evolutionary algorithms, some answers are found in scalable, classical mathematical models. Though dubbed an "AI company," our core is driving capabilities to solve problems, boost revenues, and cut costs. Our success is gauged by bettering these areas for clients.

To set the Vision on the experimentation l tend to begin with historical context of AI. I’ll say that many of my teams are surprised that it is NOT a new technology. This grounds them. Next step is to bring out why AI came out of latest winter. I’ll share that one of my clients quickly realized that they MUST include data management as part of the AI experimentation. The harder step I encounter is to convey that ML is not the whole AI. This sets a realistic expectation. For example, how do you deal with model objects in operations? In my experience I found clients who included this in the AI vision were able to go beyond experimentation

I suggest making many prototypes aimed to solve problems your company is facing. It is far easier in the GPT era. They don’t need to be fancy, just show potential- you can go a long way with good prompting and the OpenAI playground. Most prototypes will fail, but some will not. Once one or two successful prototypes are shown, people will replicate your efforts and start experimenting themselves. This will produce more successful prototypes, which will produce a virtuous cycle.

Before suggesting someone dive into artificial intelligence, they should grasp how complexity can emerge spontaneously. How something can self-design without guided intelligence. I'd first recommend reading Richard Dawkins' "The Selfish Gene" or "The Blind Watchmaker." Then, delve into numerical methods basics. This gives an initial understanding of how complex systems can arise from simple rules. Only then consider courses on PyTorch or TensorFlow.

Building expertise in AI isn't just for the tech experts; it's a team-wide endeavor. While traditional courses and tutorials offer a good starting point, real learning happens when we dive in and experiment with real projects. Everyone, whether you're in coding, design, sales, or marketing, brings a unique perspective that can make our AI initiatives more robust. By getting hands-on experience, we'll quickly adapt to AI's evolving landscape, turning our failures and successes into valuable lessons. We should provide access to tools and share findings through workshops, notes, and other means.

Two big problems with AI today are - 1. AI can lie confidently 2. AI solutions can quickly become cost prohibitive. Be mindful of these when creating teams. You must have a disciplined approach of validating and evaluating AI solutions. Ensure your team has a healthy mix of domain experts, engineers, and business people. Establish processes to make sure AI experiment goes through in-depth human validation and feedback from your domain experts. Ensure your team works with a good representation of your actual data. If you are going to use us an external AI tool then identify how much it will cost to run the solution at your scale of operations to avoid scraping the solution altogether later due to cost considerations.

Our approach to AI projects is grounded in forming cross-functional teams. Each team is comprised of members from diverse departments—be it coding, design, sales, or marketing—to ensure a balanced mix of skills and perspectives. This multifaceted team is empowered to take decisions autonomously, subject to broad project guidelines and objectives. The idea is to break down silos and foster collaboration, enabling us to more quickly adapt to AI's evolving landscape. By employing this cross-functional, decentralized approach, we aim to accelerate our learning curve, broaden the impact of our AI efforts, and produce solutions that are holistic and well-rounded.

Failures aren't setbacks; they're lessons that propel us forward. Our teams should embrace experimentation, knowing that every outcome—be it a success or a failure—equips us with insights to adapt and grow in the world of AI. This isn't just about individual learning; make it a practice to share our failures and findings with the larger team. Through workshops, notes, and other collaborative forums, so we can learn from each other.

One thing to keep in mind is that as AI integrates into operations, it automates repetitive tasks, shifting human focus. For example, chatbots address routine customer service inquiries, leading representatives to tackle more complex issues, emphasizing problem-solving over basic product knowledge. Hence, with changing roles, re-training becomes vital. Employees must understand and work effectively with AI systems, encompassing both technical know-how and soft skills, like managing evolving customer expectations. This newly reinvented process will require careful change management to deploy as well.

Gamification and recognition are good ways to encourage employees adopt AI tools. In our quarterly townhalls, we felicitate AI champions who have identified a better way of working using AI and/or have identified a value-adding AI based solution to known customer problems. On gamification side, we have our CEOs / CXOs share their AI avatars, AI generated video featuring themselves, their voice clones and other exciting possibilities and encourage employees to try something exciting themselves and share with everyone. Many a times employees are given mini-challenges that should be solved by using AI tools and the winning solution is circulated to everyone in the company.

This is probably a great way to start if you are trying to build a structured AI capability for your organization. This takes a lot of time as well as a team who can learn and start building something of their own. For experimentation and MVP (Minimum Viable Product) there are tons of tools and Services available to figure out if it will work for you. So I would say take baby steps, find out what works and then go for the big jump.

I start by educating teams on the journey AI has taken since 1950s. Many of them are surprised at the age of this elusive and hype magnetic technology. I find this first step to set a basis for the Experimentation options. Many leaders have shared positive feedback on this grounding. I follow that up with why this phase of thawing of AI winter makes sense. Obviously the progress on data availability and big data platforms spurred this phase of AI. I’ll share that one of my clients connected the dots and realized the importance of data and included some of improvements needed for data as PART of AI experimentation. A key equation I use is that ML is not the whole AI. I’ll admit that I had to build agents to showcase this paradigm.