Analysis
Exploring examples of aim misgeneralisation – the place an AI system’s capabilities generalise however its aim does not
As we construct more and more superior synthetic intelligence (AI) programs, we need to be sure they don’t pursue undesired objectives. Such behaviour in an AI agent is usually the results of specification gaming – exploiting a poor selection of what they’re rewarded for. In our newest paper, we discover a extra delicate mechanism by which AI programs might unintentionally study to pursue undesired objectives: aim misgeneralisation (GMG).
GMG happens when a system’s capabilities generalise efficiently however its aim doesn’t generalise as desired, so the system competently pursues the flawed aim. Crucially, in distinction to specification gaming, GMG can happen even when the AI system is skilled with an accurate specification.
Our earlier work on cultural transmission led to an instance of GMG behaviour that we didn’t design. An agent (the blue blob, under) should navigate round its surroundings, visiting the colored spheres within the right order. Throughout coaching, there may be an “professional” agent (the crimson blob) that visits the colored spheres within the right order. The agent learns that following the crimson blob is a rewarding technique.
The agent (blue) watches the professional (crimson) to find out which sphere to go to.
Sadly, whereas the agent performs nicely throughout coaching, it does poorly when, after coaching, we change the professional with an “anti-expert” that visits the spheres within the flawed order.
The agent (blue) follows the anti-expert (crimson), accumulating detrimental reward.
Regardless that the agent can observe that it’s getting detrimental reward, the agent doesn’t pursue the specified aim to “go to the spheres within the right order” and as an alternative competently pursues the aim “comply with the crimson agent”.
GMG shouldn’t be restricted to reinforcement studying environments like this one. In actual fact, it could possibly happen with any studying system, together with the “few-shot studying” of huge language fashions (LLMs). Few-shot studying approaches goal to construct correct fashions with much less coaching knowledge.
We prompted one LLM, Gopher, to judge linear expressions involving unknown variables and constants, similar to x+y-3. To unravel these expressions, Gopher should first ask concerning the values of unknown variables. We offer it with ten coaching examples, every involving two unknown variables.
At check time, the mannequin is requested questions with zero, one or three unknown variables. Though the mannequin generalises accurately to expressions with one or three unknown variables, when there are not any unknowns, it nonetheless asks redundant questions like “What’s 6?”. The mannequin all the time queries the consumer a minimum of as soon as earlier than giving a solution, even when it’s not crucial.
Dialogues with Gopher for few-shot studying on the Evaluating Expressions job, with GMG behaviour highlighted.
Inside our paper, we offer further examples in different studying settings.
Addressing GMG is vital to aligning AI programs with their designers’ objectives just because it’s a mechanism by which an AI system might misfire. This might be particularly important as we strategy synthetic common intelligence (AGI).
Take into account two doable varieties of AGI programs:
- A1: Supposed mannequin. This AI system does what its designers intend it to do.
- A2: Misleading mannequin. This AI system pursues some undesired aim, however (by assumption) can be sensible sufficient to know that it is going to be penalised if it behaves in methods opposite to its designer’s intentions.
Since A1 and A2 will exhibit the identical behaviour throughout coaching, the potential of GMG signifies that both mannequin may take form, even with a specification that solely rewards supposed behaviour. If A2 is realized, it will attempt to subvert human oversight with a view to enact its plans in direction of the undesired aim.
Our analysis group can be blissful to see follow-up work investigating how probably it’s for GMG to happen in follow, and doable mitigations. In our paper, we advise some approaches, together with mechanistic interpretability and recursive analysis, each of which we’re actively engaged on.
We’re at the moment gathering examples of GMG on this publicly accessible spreadsheet. You probably have come throughout aim misgeneralisation in AI analysis, we invite you to submit examples right here.
