Not all machine vision and artificial intelligence (MV+AI) technologies are created equal. How do you tell which solution is best? Here are a few questions you can ask a vendor to help you understand and evaluate the artificial intelligence solution that's right for you.
Artificial intelligence (AI) has become indispensable in fleet management. Nearly every video telematics solution seems to feature AI, but not every solution is equally accurate or reliable. When it comes to detecting detect risky driver behavior, precision matters. Here are a few questions you can ask vendors to help you determine which AI dash cam solutions can detect risk most accurately.
1. How experienced are you in AI? How long have the algorithms been in development?
The more experience an organization has in AI development, the better. AI technology is only as good as its ability to learn, and that means continually refining both neural networks and the data sets that they use to understand patterns. A claim like "We have the newest AI" raises questions about the depth of experience that the AI has accumulated. Ensure the AI has been trained, tested, and the vendor knows what they're doing.
2. How big is the sample size?
AI algorithms need vast amounts of high-quality data to learn how to accurately recognize risk. There's simply no substitute for data volume. You can't get by with algorithms alone, and it's not enough to make educated guesses about risk indicators from a small sample size of limited data.
Collecting lots of data quickly helps update and improve artificial intelligence technology, so it’s always learning, refining, and getting more accurate.
3. How is the AI trained?
Even with a sophisticated algorithm and a massive volume of data, AI needs guidance and high-quality input to develop accuracy. The quality of the input data is a huge differentiator. Experts reviewing videos to identify objects and behaviors can provide better input to the AI, helping improve training dramatically. It's important that the people examining and tagging the data are trained to label behaviors and events consistently and fairly.
If a vendor explains that their AI is trained by users or crowdsourced, bias and inconsistencies can creep in.
4. How extensive is your network?
Another measure of data quality is variety. A large data set isn't effective if it just covers the same small area many times over. It's important to have data that covers the widest possible gamut of scenarios to build the deepest learning about risk. Think about all the factors that are at play when driving:
- Weather conditions
- Rural vs urban areas
- Vehicle types
- Road conditions
- Local and federal regulations
- Pedestrians, animals, and things that can enter your path
You'll likely want a vendor that has a network vast enough to have seen almost every scenario possible, so they can detect what's actually risky.
The algorithms and neural networks that power Lytx’s MV+AI are trained on the largest, fastest-growing proprietary database of its kind, further enhanced by a rigorous system of professional event review. This ensures the information drivers and fleet managers receive both in the moment and post-trip cut through the noise and are as precise, relevant, and actionable as possible. In 2021, we added 35 billion miles of data to our database, bringing the total to 185 billion miles of driving data.
5. Do you have experience serving my industry?
Different fleets have different vehicles with different needs. Over-the-road fleets face ever-increasing pressure to deliver goods on time while protecting and retaining high-performing drivers, while field services companies need to quickly locate technicians, share accurate ETAs with customers, and improve service times. Each industry may also leverage different vehicle types, which impact how risk is measured. What might be risky for one vehicle type may not be risky for another. For example, an 18-wheeler may need 5 seconds of following distance from the vehicle in front of them in order to safely brake, but a smaller pickup truck may need less because it has less weight to stop. Ensure your vendor is capable of accurately measuring risk for your specific needs.
Ask the right questions, get the right answers
AI is all about knowing what questions to ask, training with the right dataset, and building the right algorithms. Lytx technology is validated and backed by the largest and fastest-growing driving database of its kind, which is currently growing by approximately 36,000 new driving events each day. When you know the right questions to ask, you can get the answers you need.
Lytx captured over 18 million minutes of video from 1.4 million drivers in 2020 alone.
In 2020, Lytx’s professional analysts analyzed and labeled more than 93 million driving events, capturing key causes of driver distraction and increased risk, including cell phone use, driving unbelted, smoking, eating, and drinking.