A Fruitful Conversation
Do you ever talk to your plants? Now they can talk back with the help of a Raspberry Pi, environmental sensors, and an AI chatbot.
When we see a person conversing with an inanimate object, a plant, or anything else that is incapable of responding, our first thought may be that they should seek some type of therapy to restore their grip on reality. But that knee-jerk reaction may no longer be an appropriate one. All manner of things, both living and nonliving, may soon acquire the power of speech with the help of modern technology. Needs or desires that cannot be vocalized can now be made known through plain language.
Creative technologist Matt Reed of redpepper recently described how conversational skills were given to tomato plants, of all things, using the power of artificial intelligence. At first blush, this sounds like little more than a gimmick, but that is not entirely true in this case. It is true that a tomato plant has no ability to reason or express its needs to us, but nevertheless, it does have needs that can be known.
Plants react in response to factors like environmental conditions, stresses, or touch, and these variables can be measured with sensors. Leveraging these observations collected from sensors, in conjunction with the knowledge about the world that is contained in a large language model (LLM) and a few other tricks, a sensible conversation can be carried on between a person and a plant. These conversations can uncover useful information, like if the plant needs to be watered or could use a bit more sunshine. Sure, the raw sensor readings could also reveal these needs, but simply asking the plant what it needs is a lot more intuitive. Well, it is intuitive after you get used to the idea of talking to plants, anyway.
The initial prototype makes use of a Raspberry Pi single-board computer to handle all of the system’s processing needs. For starters, soil moisture and light sensors were hooked up to the Raspberry Pi. A customized prompt is fed into OpenAI's GPT-4 LLM telling it to act like a tomato plant, along with readings from the sensors. Then when someone talks, the speech is converted to text and sent to the LLM. The LLM responds in a sensible way (hopefully!), given the initial prompting and the user’s query. Finally, the response from GPT-4 is converted to speech and played audibly.
This project may be giving you some serious Little Shop of Horrors vibes, but Reed assures us that the system will never ask for anything more than water or sunlight. Fear not, there is nothing evil hidden behind those googly eyes stuck to the hardware.
Looking to the future, Reed envisions adding some richer sensing capabilities to the plant. Perhaps by adding air quality, pheromone, and soil nutrient sensors, and so on, richer interactions could be achieved. And if a tomato plant can talk, why not other things as well? What would you like to give a voice to?
