Artificial intelligence Cybersecurity Health, Wellness & Biotech SaaS Transportation & Logistics Workplace

5 Key Areas For ‘Hard-Task’ AI To Conquer

Illustration of head with 1 and 0 code brain/data

By Bob Rosin

We hurtle down the highway, well above the speed limit, moving easily over the smooth road, trusting our Tesla’s Autopilot to handle the slight bends as they come.

But as we enter the Golden Gate Bridge heading toward San Francisco, the lane narrows, encroached on by construction and temporary barriers on the left, the road becomes unpredictable, uneven and full of cracks — I immediately take control of the wheel.

Bob Rosin, partner at Defy
Bob Rosin, partner at

Anyone who has tried Autopilot in a Tesla is sure to have had a similar experience. It’s good enough to handle the monotonous long highway stretches. But when the driving gets challenging, it’s clearly not up to the task.

But what’s wrong with this picture? Is there some inherent reason why AI should be good at the easy, mindless tasks, but bad at the hard ones? Logically, as processing power increases and models improve, at some threshold Autopilot ought to be better than a human.

A few years from now, when the road narrows and the driving gets challenging, I expect to turn Autopilot on.

Easy-task AI

If we look at the state of AI companies today, we find parallels. With the world agog at the surprising abilities of ChatGPT, new companies pop up every day purportedly serving novel cases across every industry.

For the most part, these use cases are analogous to driving along a smooth, paved highway. They fall primarily into “easy-task AI” scenarios: high volume, low criticality and, relative to other scenarios, high tolerance for error. Most AI-based tools today focus on repetitive tasks, letting humans address the high-value, mission-critical situations.

Some examples: More than $1 billion in venture capital has been raised for AI startups in customer service, including Uniphore, Forethought, Moveworks, Observe.AI and Gorgias, in addition to products from incumbents such as Intercom’s AI bot, Fin.

Enterprises may tolerate errors in support workflows, provided a large volume of inquiries can be addressed cost-effectively; customers will always escalate to a human if the AI doesn’t get the job done.

The state of AI in legal is similar: DoNotPay is a brilliant example of using AI to brute force solve legal issues where it’s OK to sometimes be wrong, such as fighting parking tickets, canceling subscriptions and myriad other fairly low-stakes legal tasks.

Similarly in medicine, companies like Abridge save time by automating clinical notes. In this case, AI is not replacing doctors, just making them more efficient.

But is that the future? It seems odd to relegate AI models — trained on more data than an individual could ever internalize — to only relieving humans of tasks that most would consider repetitive, mundane or “easy.”

Instead, it seems far more likely that AI will begin to take on the hard tasks.

Hard-task AI

What does the world look like when the most difficult, mission-critical, high-risk tasks are the ones that AI does best?

We are seeing hints already. Researchers from MIT and Massachusetts General Hospital are developing an AI model that analyzes CT scans, potentially detecting lung cancer years earlier than a human radiologist. Is there a day in the future that a bot with a vast corpus of knowledge of previous cases will be your primary care physician?

The U.S. Air Force has demonstrated the X-62A Vista, an AI-piloted fighter jet with faster response times and greater precision than a human pilot. portfolio company is building a virtual sales engineer: Imagine if a sales rep on their first day already knows how to flawlessly answer all the questions that would have required a product expert or SE to be on the call? Would you ever do a high-stakes sales call without your AI assistant?

As AI takes on tasks beyond the capabilities of humans, here are some industries we can expect to be impacted:

Medicine: AI systems that analyze clinical data and predict diagnoses with more precision than human doctors, and provide recommendations for medication and treatment plans. AI-assisted surgery is in its infancy. Drug discovery is already being revolutionized by AI.

Transportation: Autonomous vehicles are only the beginning. Beyond platooning of trucks, imagine if vehicles on the road communicate with one another and form a network, acting effectively as a single organism to adaptively minimize congestion and operate at higher speeds safely, rather than compounding delays as each driver responds.

Enterprise SaaS: Why must every CIO reinvent the wheel in their organizations? Internal systems will be self-integrating; automation will connect systems from disparate vendors to achieve complex tasks. Intelligent analysis of data will be through conversational interfaces.

Security: Advanced models already detect fraud by analyzing patterns across millions of transactions, far beyond the skills of any human. Sophisticated AI systems are already scoring risk across hundreds of thousands of employees. We will live in a future without passwords, where enterprise security systems operate silently behind the scenes adapting to signals from a shared security network across enterprises.

Workforce of the future: The questions are inevitable. What is the role of humans in this new world? What are the interfaces between humans and the new intelligent systems? What governance models are needed? How much autonomy do we accord these systems? What skills should we be teaching our children to prepare for a future where the hard intellectual challenges are handled by intelligent systems?

I’d like to hear what use cases for AI you are envisioning for the future.

Bob Rosin is an investment partner at As a founder, serial entrepreneur and former leadership team member at Skype, LinkedIn and Stripe, he’s experienced all facets of startup life. Rosin serves on the boards of GajiGesa, Elevate Security and Aircover. He’s also an active angel investor and adviser to companies including, among others, Stripe, Workato, Tenor (acquired by Google), Cursor Data (acquired by DataRobot), MindMeld (acquired by Cisco), Instawork, Tonal Fitness, Accord and Prairie Health.

Illustration: Dom Guzman

Search less. Close more.

Grow your revenue with all-in-one prospecting solutions powered by the leader in private-company data.

Stay up to date with recent funding rounds, acquisitions, and more with the Crunchbase Daily.

Copy link