I always had the informal impression that the optimal policies were deterministic (choosing the best option, rather than some mix of options). Of course, this is not the case when facing other agents, but I had the impression this would hold when facing the environment rather that other players.

But stochastic policies can also be needed if the environment is partially observable, at least if the policy is Markov (memoryless). Consider the following POMDP (partially observable Markov decision process):

There are two states, 1a and 1b, and the agent cannot tell which one they're in. Action A in state 1a and B in state 1b, gives a reward of -R and keeps the agent in the same place. Action B in state 1a and A in state 1b, gives a reward of R and moves the agent to the other state.

The returns for the two deterministic policies - A and B - are -R every turn except maybe for the first. While the return for the stochastic policy of 0.5A + 0.5B is 0 per turn.

Of course, if the agent can observe the reward, the environment is no longer partially observable (though we can imagine the reward is delayed until later). And the general policy of "alternate A and B" is more effective that the 0.5A + 0.5B policy. Still, that stochastic policy is the best of the memoryless policies available in this POMDP.

From the Google Research blog:

We believe that AI technologies are likely to be overwhelmingly useful and beneficial for humanity. But part of being a responsible steward of any new technology is thinking through potential challenges and how best to address any associated risks. So today we’re publishing a technical paper, Concrete Problems in AI Safety, a collaboration among scientists at Google, OpenAI, Stanford and Berkeley.

While possible AI safety risks have received a lot of public attention, most previous discussion has been very hypothetical and speculative. We believe it’s essential to ground concerns in real machine learning research, and to start developing practical approaches for engineering AI systems that operate safely and reliably.

We’ve outlined five problems we think will be very important as we apply AI in more general circumstances. These are all forward thinking, long-term research questions -- minor issues today, but important to address for future systems:

• Avoiding Negative Side Effects: How can we ensure that an AI system will not disturb its environment in negative ways while pursuing its goals, e.g. a cleaning robot knocking over a vase because it can clean faster by doing so?
• Avoiding Reward Hacking: How can we avoid gaming of the reward function? For example, we don’t want this cleaning robot simply covering over messes with materials it can’t see through.
• Scalable Oversight: How can we efficiently ensure that a given AI system respects aspects of the objective that are too expensive to be frequently evaluated during training? For example, if an AI system gets human feedback as it performs a task, it needs to use that feedback efficiently because asking too often would be annoying.
• Safe Exploration: How do we ensure that an AI system doesn’t make exploratory moves with very negative repercussions? For example, maybe a cleaning robot should experiment with mopping strategies, but clearly it shouldn’t try putting a wet mop in an electrical outlet.
• Robustness to Distributional Shift: How do we ensure that an AI system recognizes, and behaves robustly, when it’s in an environment very different from its training environment? For example, heuristics learned for a factory workfloor may not be safe enough for an office.

We go into more technical detail in the paper. The machine learning research community has already thought quite a bit about most of these problems and many related issues, but we think there’s a lot more work to be done.

We believe in rigorous, open, cross-institution work on how to build machine learning systems that work as intended. We’re eager to continue our collaborations with other research groups to make positive progress on AI.

Many people are probably aware of the hack at DAO, using a bug in their smart contract system to steal millions of dollars worth of the crypto currency Ethereum.

There's various arguments as to whether this theft was technically allowed or not, and what should be done about it, and so on. Many people are arguing that the code is the contract, and that therefore no-one should be allowed to interfere with it - DAO just made a coding mistake, and are now being (deservedly?) punished for it.

That got me wondering whether its ever possible to make a smart contract without a full AI of some sort. For instance, if the contract is triggered by the delivery of physical goods - how can you define what the goods are, what constitutes delivery, what constitutes possession of them, and so on. You could have a human confirm delivery - but that's precisely the kind of judgement call you want to avoid. You could have an automated delivery confirmation system - but what happens if someone hacks or triggers that? You could connect it automatically with scanning headlines of media reports, but again, this is relying on aggregated human judgement, which could be hacked or influenced.

Digital goods seem more secure, as you can automate confirmation of delivery/services rendered, and so on. But, again, this leaves the confirmation process open to hacking. Which would be illegal, if you're going to profit from the hack. Hum...

This seems the most promising avenue for smart contracts that doesn't involve full AI: clear out the bugs in the code, then ground the confirmation procedure in such a way that it can only be hacked in a way that's already illegal. Sort of use the standard legal system as a backstop, fixing the basic assumptions, and then setting up the smart contracts on top of them (which is not the same as using the standard legal system within the contract).