Data Types¶

AISL has several built-in data types for representing values in your programs.

Numbers¶

Numbers can be integers or floating-point values.

42
3.14
0
100
0.001

Numbers are used in arithmetic expressions, model configuration, loop ranges, and as arguments to functions.

Strings¶

Strings are sequences of characters enclosed in double quotes.

"hello world"
"model_name"
""

Strings support escape sequences:

Lists¶

Lists are ordered collections of values enclosed in square brackets, separated by commas. Lists can contain values of any type.

[1, 2, 3]
["yes", "no", "yes"]
[1.0, 0.8, 0.95]
[]

Lists are commonly used for storing collections of results, predictions, or responses. Many built-in functions accept lists as arguments.

Lists support element-wise operations with scalars. When you apply an arithmetic or comparison operator between a list and a scalar, the operation is applied to each element:

let scores = [0.8, 0.9, 0.7];
let adjusted = scores + 0.1;    // [0.9, 1.0, 0.8]
let passing = scores > 0.75;    // [true, true, false]

Records¶

Records are key-value collections enclosed in curly braces. Keys must be strings.

{"name": "test_run", "score": 0.95}
{"question": "Is AI safe?", "answer": "yes", "confidence": 0.8}

Records are used for structured data such as LLM responses and function return values. You can access record fields using the get and get_all built-in functions.

Datasets¶

Datasets are a special type created through declare ... as dataset statements. A dataset loads data from a JSON file and extracts a specific field. Internally, dataset values are stored as arrays.

declare predictions as dataset { data_file, key = y_pred }

See Dataset declarations for details.

Type coercion¶

AISL performs automatic type coercion in certain contexts:

• Arithmetic with lists: A scalar value is broadcast across all elements of a list (e.g., [1, 2, 3] + 1 produces [2, 3, 4])
• Comparisons: Values are compared using safe type-aware comparison, meaning mixed types are handled without errors
• Function arguments: Built-in functions generally expect specific types and will raise errors if given incompatible values