Open AI and Chat GPT are at L2 level of Artificial Intelligence, but going on L3
The framework of levels of automated analytics & action on data published last week is helpful for assessing how far any AI product has come on the path to automation. See here for a reminder.
Even though the framework was meant for assessing analytics tools, it actually can be applied much broader. That’s because, in a sense, any input into an AI is a type of “data.” That data is analyzed by the AI and turned into an output.
That output can just be content that is generated, which is how we know Gen AI today. Or, the output can be connected with tools that can take actions, e.g. similar to how the driver assistance AI in a self-driving car can control the steering wheel.
Open AI is at Level 2 today
As every kid knows by now, Chat GPT can be used to generate, summarize, and translate content. It can also interpret tabular data. So, to use the example for Level 2 in the diagram above, if provided with the inputs of a customer survey along with behavioral data from those customers within a table, it can be used to extract a correlation.
What does it mean to cross from Level 2 to Level 3?
The cross-over into Level 3 is a big step, not at all trivial. As a reminder, the leveling hierarchy here is inspired by the SAE’s 6 levels of autonomous driving. And today, there is only a single car manufacturer that has dipped their toes into getting certified for Level 3, and only in very limited regions.
- A fully Level 3 certified vehicle could for example serve as an autonomous local taxi, according to the SAE’s hierarchy
- We are still far, far away from Level 5 where a car can drive autonomously “everywhere in all conditions”
Similarly, for the levels of automated analytics & action, Level 3 requires automating tasks within a limited scope. In the table above, for example:
- A Level 3 analytics product could auto generate A/B tests for a website and run these tests automatically with a pre-defined “safe” audience.
- A level 5 analytics product would run your digital channel for you. We are still far, far away from that.
Open AI is crossing into Level 3
Already today, applications can be automated using Open AI’s APIs, e.g. for limited scope use cases such as the following:
- Contact center chatbot automation for the more simple part of the conversation, for example, see a case study here.
- Automated completion of form entries by extracting the main takeaways from a given text or conversation.
For example, an automation application built with Chat GPT might be more reliable for keeping CRM records updated based on the latest customer email and phone conversations than human sales people typically are today.
Additionally, though, Open AI just announced that they are developing software that operates devices, automates tasks,
There you have it, that would bring Open AI products themselves to level 3 where tasks are automated within a limited scope.
What would it mean for Open AI to reach Level 4 or 5?
At level 5, general artifical intelligence, AI could replace humans for desk work, just like self-driving cars at Level 5 could drive “everywhere in all conditions”, no steering wheel or pedals for humans required. But we are still far far away from that.
Could an AI at Level 5 cook dinner for your family though?
That would require a degree of physical, robotic automation and nimbleness that we haven’t seen yet outside of Sci Fi movies.
But … looking at the news on February 29th, this is exactly what’s ahead now. Humanoid robot startup Figure AI valued at $2.6 billion as Bezos, OpenAI, Nvidia join funding
Image embedded from CNBC’s article.
And it seems plausible too. If AI can drive a car, surely we can get to where it can control a robot. There is no reason for these robots to look like a human though, except if it makes us humans feel more comfortable around them.
Does it though?
Leave a Reply