Top 3 Challenges to achieving “AI Nirvana”
Updated: Feb 17
Having spent over two decades applying AI and cognitive technologies - from early-stage startups to Fortune 50 companies - I have identified three key challenges to achieving “AI Nirvana”.
Let’s face it. AI is on everyone’s mind these days.
Companies in all categories seek to enhance their product value by applying AI to vast troves of digital data that they have captured in data warehouses.
Infinitely elastic cloud resources are available for rent. Customers want to increase their operational productivity. They seek to use AI to create smarter services to extract actionable insights from their data.
A common question is how to be strategic about applying cognitive technologies.
Why? Talent is scarce, technologies change rapidly, and in-house science projects often fail to create monetizable products.
So what’s holding us back today? Here are my top 3 insights.
Challenge 1: It’s about the data.
Machine Learning (ML) is the dominant form of AI technologies today. Advances in cognitive approaches such as Deep Learning paired with cloud computing have democratized AI.
However, no AI/ML is possible without data.
To build an AI strategy, companies must have convincing answers to these questions:
Do you own or have access to high quality training data to build predictive models relevant to your business strategy?
How will you deliver insights generated by the AI - the output of the model - to the users of those insights to maximize effectiveness?
How can you create a virtuous cycle and ensure that your AI improves with usage and installed base? Closing the feedback loop is an essential part of your workflow.
AI Feedback Cycle
Challenge 2: Selecting the right approach to applying AI/ML.
The AI field is advancing rapidly. No one can keep up with the latest advances. Furthermore, it is not obvious which approach to use for a particular situation. Even 20 years of experience architecting and developing AI/ML systems in a constantly changing field does not yield a fixed heuristic set of solutions.
There are many ways of doing exactly the same thing. E.g., should you really be leveraging an Open AI GPT-2 model for your NLP (Natural Language Processing) work, when you could make a lot of progress with a bag-of-words approach, or perhaps by using ElasticSearch to build a recommendation engine?
While there are many different AI approaches to address a particular problem, also many different domains are clubbed together as “Machine Learning”. While Convolutional Neural Networks (CNNs) are a great ML approach to use in Image Understanding, good old Support Vector Machines (SVNs) could efficiently solve a whole class of problems.
The list of options for your situation is endless. Knowing which one to use is really hard. It is worthwhile to work with experts to help make the choice as you build up a strong AI competency within your dev team.
Overview of AI algorithms
Image courtesy: Jason Brownlee
Challenge 3: The talent gap.
Lets face it, AI/ML talent is super hard to find.
But there is a silver lining. Most companies do NOT need AI specialists full-time. The day-to-day grunt work for an AI-enhanced product looks exactly the same as that for any other product: Agile process, Daily scrums, and Continuous Integration/ Continuous Delivery (CI/CD). But you do need to ensure that you are executing on the "right" AI strategy.
But you do need to ensure that you are executing on the “right” AI strategy.
Economically engage the AI workforce:
Engage specialists for selecting an approach, defining your architecture and transferring that knowledge to your team.
Set time aside to retrain your existing team in the fundamentals of AI/ML and on the specific approach you have chosen.
Hold hackathons aimed at a quick prototype implementation of your approach - not to build the final product in a weekend, but to establish the viability of the approach and to shake out the issues you will need to address for product rollout.
The above challenges apply even to companies that are already on the journey of building an AI /ML infused product. It’s important to do periodic assessments of whether your data is being used adequately, whether your approach is still relevant or has been supplanted by a recent advance and whether your talent is up to the challenge.
What’s next: Better resources are coming to help achieve AI nirvana.
In collaboration with my colleagues at WorksMachine, I am developing a playbook for companies to integrate AI into their product offering.
Our methodology will combine best practices in product, AI, and category strategy to help your company get to the next level.
Partner | Product & AI Strategy