Делаем простой ИИ тамагочи на ChatGPT

Обсуждали с коллегами что ещё может современный ИИ, вспомнили про тамагочи и всякую искусственную жизнь.

Поболтали и разошлись, а я пошёл и за 15 минут получил вот такой результат.

Тюнить и украшать вывод не буду — уже есть куча примеров как получать чистый код от ботов. В данном случае интересна сама демонстрация возможности.

Спойлер: с ChatGPT не надо тратить время на разработку ИИ простых ботов, сеть подстроится под ваши требования. Просто скажите кем ChatGPT должна притвориться.

Этот пост неявно продолжает размышления из Write Your Own Adventure.

Простая демонстрация

Tiendil

Let’s imagine a simplified virtual world with the following:

- A hamster — autonomous agent with the behaviour expected from a normal hamster.
- A cell in which this hamster lives.
- Hamster stays in the centre of the cell.
- From the left side of the cell, there is cave A.
- From the right side of the cell, there is cave B.
- The red cube blocks entry to cave A.
- The blue cube blocks entry to cave B.
- There is a mouse pointer which is visible for the hamster.
- Hamster knew that if the pointer is moving in circles, then the food is hidden in cave A.
- Hamster knew that if the pointer is moving in crosses, then the food is hiden in cave B.
- The hamster can easily move blocks.

We can describe hamster actions with the next notation:

- Terms are `hamster`, `cave_a`, `cave_b`, `cube_red`, `cube_blue`, `food`
- Operations are `go_to(X, Y)`, `move_away(x, Y)`, `eat(X, Y)`
- Full operation look like `go_to(hamster, cube_red)` or `eats(hamster, food)`.

Describe the next actions of the hamster for the specified situation in format:

- `operation 1`
- `operation 2`
- …
- `operation n`

ChatGPT

Here are the steps the hamster could take to get the food in this situation:


go_to(hamster, cube_red) # Move to the red cube
move_away(cube_red, hamster) # Push the red cube away
go_to(hamster, cave_a) # Move to cave A
eat(hamster, food) # Eat the food in cave A


Alternatively, if the mouse pointer is moving in crosses instead of circles, the hamster would do the following:


go_to(hamster, cube_blue) # Move to the blue cube
move_away(cube_blue, hamster) # Push the blue cube away
go_to(hamster, cave_b) # Move to cave B
eat(hamster, food) # Eat the food in cave B

Сложное поведение

Tiendil

Let’s imagine a simplified virtual world with the following:

- A hamster — autonomous agent with the behaviour expected from a normal hamster.
- A cell in which this hamster lives.
- Hamster stays in the centre of the cell.
- From the west side of the cell, there is cave A.
- From the east side of the cell, there is cave B.
- The red cube blocks entry to cave A.
- The blue cube blocks entry to cave B.
- The pink button is placed near north side of the cell.
- There is a mouse pointer which is visible for the hamster.
- The hamster can easily move cubes if they activated.
- Red cube is not activated.
- Blue cube is not activated.
- Hamster must push pink button 3 times to activste red cube.
- Hamster must push pink button 5 times to activste blue cube.
- Hamster knew that if the pointer is moving in circles, then the food is hidden in cave A.
- Hamster knew that if the pointer is moving in crosses, then the food is hiden in cave B.

We can describe hamster actions with the next notation:

- Terms are `hamster`, `cave_a`, `cave_b`, `cube_red`, `cube_blue`, `food`, `button`
- Operations are `go_to(X, Y)`, `move_away(X, Y)`, `eat(X, Y)`, `push(X, Y)`.
- Full operation look like `go_to(hamster, cube_red)` or `eats(hamster, food)`.

Describe the next actions of the hamster for the specified situation in format:

- `operation 1`
- `operation 2`
- …
- `operation n`

ChatGPT

Here is a possible set of actions for the hamster in this situation:

go_to(hamster, button): Move to the pink button.
push(hamster, button): Push the button once.
push(hamster, button): Push the button twice.
push(hamster, button): Push the button three times to activate the red cube.
go_to(hamster, cave_a): Move to the entrance of cave A.
go_to(hamster, cube_red): Move to the red cube.
move_away(hamster, cube_red): Push the red cube out of the way.
go_to(hamster, food): Move to the food in cave A.
eat(hamster, food): Eat the food.
go_to(hamster, cell): Move back to the center of the cell.
go_to(hamster, button): Move to the pink button.
push(hamster, button): Push the button four times.
push(hamster, button): Push the button five times to activate the blue cube.
go_to(hamster, cave_b): Move to the entrance of cave B.
go_to(hamster, cube_blue): Move to the blue cube.
move_away(hamster, cube_blue): Push the blue cube out of the way.
go_to(hamster, food): Move to the food in cave B.
eat(hamster, food): Eat the food.
go_to(hamster, cell): Move back to the center of the cell.

Тут стоит обратить внимание на пару нюансов:

• Это уже довольно сложное поведение.
• Активируя второй куб ChatGPT учла сколько раз была нажата кнопка ранее.
• ChatGPT в этот раз не указала на что реагировал хомяк, просто открыла обе пещеры. Но тут я виноват, что не прописал явно как двигается курсор.

Первый вариант запроса

Самый первый запрос дал немного другие результаты, но тоже интересные.

Tiendil

Let’s imagine a simplified virtual world with the following:

- A hamster — autonomous agent with the behaviour expected from a normal hamster.
- A cell in which this hamster lives.
- Hamster stays in the centre of the cell.
- From the left side of the cell, there is cave A.
- From the right side of the cell, there is cave B.
- The red cube blocks entry to cave A.
- The blue cube blocks entry to cave B.
- There is a mouse pointer which is visible for the hamster.
- Hamster knew that if the pointer is moving in circles, then the food is hidden in cave A.
- Hamster knew that if the pointer is moving in crosses, then the food is hiden in cave B.
- The hamster can easily move blocks.

Describe in terms of logic predicates what the hamster will do if he is hungry and sees that pointer moved in crosses.

ChatGPT

To describe the behavior of the hamster in this scenario, we can use logical predicates that capture the rules and conditions of the virtual world.

Let’s use the following predicates:

- `hungry(X)`: True if X is the hamster and X is hungry.
- \`in\_cell(X): True if X is inside the cell.
- `can_move(X, Y)`: True if X can move Y (where Y is the object being moved).
- `blocked(X, Y)`: True if object Y is blocking access to object X.
- `sees(X, Y)`: True if X can see object Y.
- `moving_in_crosses(X)`: True if the pointer is moving in crosses.
- `food_in_cave(X, Y)`: True if food is in cave Y and hamster X knows it.

With these predicates, we can describe the behavior of the hamster as follows:

- If the hamster is hungry and sees the pointer moving in crosses, it will first check if it is inside the cell using the in\_cell predicate.
- If it is inside the cell, it will check if it can move the blue cube blocking cave B using the can\_move predicate.
- If it can move the blue cube, it will move it out of the way using the can\_move predicate.
- It will then check if it can see the food in cave B using the sees predicate.
- If it can see the food, it will go to cave B to get it.
- If it cannot see the food, it will assume that the food is in cave A (because it knows that the pointer moving in crosses means the food is in cave B) and check if it can move the red cube blocking cave A using the can\_move predicate.
- If it can move the red cube, it will move it out of the way using the can\_move predicate.
- It will then check if it can see the food in cave A using the sees predicate.
- If it can see the food, it will go to cave A to get it.
- If it cannot see the food in either cave, it will assume that it made an error in its inference and try again later.
- Therefore, the logical statement that describes the behavior of the hamster when it is hungry and sees the pointer moving in crosses would be:

hungry(hamster) ∧ in_cell(hamster) ∧ moving_in_crosses(pointer) ∧ can_move(hamster, blue_cube) ∧ sees(hamster, food_in_cave(hamster, cave_B))
→ (can_move(blue_cube, empty_space) ∧ sees(hamster, food_in_cave(hamster, cave_B)) ∧ go_to(cave_B))
∨ (can_move(hamster, red_cube) ∧ can_move(red_cube, empty_space) ∧ sees(hamster, food_in_cave(hamster, cave_A)) ∧ go_to(cave_A))


where `empty_space` refers to an empty space where the cube was moved and `go_to` refers to the hamster’s action of moving to the specified cave.