Key insights that will help you get the most out of AI.
It’s easy to get caught up in the hype surrounding artificial intelligence (AI), but it’s also easy to be fooled into thinking it’s all hype. The truth is somewhere in the middle.Or, as tech celebrity Mike Olson Suggest“The stifling focus on AGI and self-driving cars and blinds like that [us] Focus on the value of narrow AI applications. “Narrowly focused,” he was referring to DeepMind’s announcement that it had published “predicted structures for nearly all cataloged proteins known to science.”
narrow? almost not. This advance has greatly opened up access to protein structures, thereby accelerating scientific discoveries in diverse fields such as medicine and climate change. But the AI used Yes In a sense, it’s not some kind of sentient machine that thinks through protein structures. As I’ve written, the best machine learning (ML) is often “just” pattern matching on a scale that humans cannot replicate.
Consider this a reminder that just because AI/ML doesn’t give us the flair for self-driving cars doesn’t mean it hasn’t made impressive strides. The trick is to narrow the scope of our use of AI, rather than abandon its promise.
The right way to do AI is to use machines for what they are good at and supplement that with human intelligence. Machines can process vast amounts of information, far beyond any human capability, and then present that information to people in a way that is easier for them to understand and hypothesize. This is not a human battle. Machines – Humans working with machines.
ocean: AI Ethics Policy (Tech Republic Premium)
And data. a lot of. In fact, no matter how good the machines are and how smart the people are, without data, the mapping of all known proteins would be impossible, as Ewan Birney, Deputy Director General of EMBL, said, Regulation. “All the AI talent in the world…it’s not easy[ily] Solving scientific problems…no data — and there’s a lot more. So where did the DeepMind scientists get this data? Fortunately, in this particular field, there is a tradition of shared data, as Birney continues: “Here, the long-established shared data in molecular biology Community norms – especially structural biology here – are a key enabler. “
Applied to a data science project within any given organization, this requires machines operating at scale, savvy data scientists, and large amounts of data. When these three things come together, AI has the potential to become truly magical, albeit not in some “perceptual machine” way, as mentioned earlier. It is still critical to point the model to relatively “narrow” problems where the machine’s advantage can be exploited, such as pattern matching.
Also, as emphasized by Aible CEO Arijit Sengupta, data scientists should be pragmatic about their models. Sengupta regularly hosts a competition pitting high school students against Berkeley-trained data science students. High school students almost always beat college students, he said, for the same reasons that most enterprise AI projects fail: “Data scientists and machine learning engineers are taught to look at ‘model performance’ (a given algorithm is the same as “in a competition, high school students Better focus on model dollar and cent results, while college students “focus on training fancy algorithms.”
In other words, it pays to keep things simple. And focus on areas where its strength is growing.
ocean: Hiring Toolkit: AI Architect (Tech Republic)
So where should businesses use AI in the near term? Since 2016, we have made “remarkable progress” in artificial intelligence, with AI showing particular progress in three key areas, according to Stanford University’s “Centennial Study of Artificial Intelligence”:
- Learn in a self-supervised or self-motivated way
- Learn in a continuous manner to solve problems in many different domains without extensive retraining in each domain
- Generalization between tasks – adapting the knowledge and skills acquired by the system for a task to new situations
Considering these parameters, businesses can go from “mostly failing” to “mostly succeeding” with AI. It’s just a matter of using AI wisely.
Disclosure: I work for MongoDB, but the opinions expressed here are mine.