SMT: Oracle Decay Problem
Ideally, oracles would have an incentive to find decentralized sources for the feeds they provide. They should be paid to have unique ways of gathering information. But it's difficult to identify the difference between clever sourcing and faked feeds. It seems like there should be watchdog oracles that can do a meta-analysis of the feeds, to validate them. Something like a sigma statistical significance.
In specific fields such as particle physics and manufacturing, statistical significance is often expressed in multiples of the standard deviation or sigma (σ) of a normal distribution, with significance thresholds set at a much stricter level (e.g. 5σ).
At some point, real-world information feeds will break down. Imagine the scenario of SMT Oracles providing real-world information feeds for specific solutions. For example, several oracles start to provide feeds for satellite information they gather. The information they provide to the blockchain includes things like the temperature, geological particulars, and elevation of a specific area.
It's a decentralized blockchain with decentralized feeds of centralized data.
If the raw data provider ever interrupts their service, all of the feeds are interrupted right along with it because none of the oracles operate the satellites. Instead, they subscribe to the information feed and pay a monthly fee, or something. Several of them do this, so it's decentralized, in a way. But the reality is, the information source is centralized. It's repeated to the blockchain and appears decentralized because the oracles all run their feed updates at different times.
That's a single point of failure that needs to be identified before business rules are designed on top of them. There should be an incentive to manage proper feeds and a disincentive to provide garbage. But a single garbage feed might not be enough of a worry for an oracle to think about.
I think the net result would be slow decay of confidence in all feeds unless some compensating control is considered.
wow i didnt realize how substantial SMTs are, i bet SMts will create a lot of new jobs!
and I can' wait for the smart forks!
smart media tokens might attract some BIG companies
Good points to think about and learn from as always. Agree as one of the things I mentions is times we have this decentralized thing. But many things that interact with it are still centralized. Seems we all have more learning to do.
Yeah, this can even be done off-chain now by looking at STEEM Witness price feeds. If everyone is just using the same script and the same exchange, it's effectively centralized.
Some thoughts:
Compensation control can come from potential loss of job - similar to block production in DPoS Witnesses.
Others thoughts for thought:
Oracles data may need to overlap at some high threshhold, such as 75%, for the data to become consensus over allowing rewards distributions.
Potential Oracles need to compete with each other for jobs and exposure of poor performance is in the challengers' interests.
Consensus over permitting rewards distributions may come from anywhere between one and thirty elected Oracles. The number of Oracles is defined by the SMT creator.
Oracles as they affect SMTs Rewards Pools are only here to answer a simple question: Did the entity behind this account perform action X, or X+Y, or X+Y+N? If yes, publish the account on the Oracle's feed.
Oracles as they affect SMTs Rewards Pools are only here to answer a simple question: Did the entity behind this account perform action X, or X+Y, or X+Y+N? If yes, publish the account on the Oracle's feed.
Some Oracle systems will fail, some will succeed. 90 to 10.
Yep, I agree that potential loss of job is the ultimate result of doing a bad job. But I'm thinking about what happens leading up to that. For instance, I don't pull a vote from a witness that misses blocks. It's a clue about what state their witness is in, though. It might be a good indicator to kick off a meta-analysis.
It's this "in-between" state that's interesting. It seems like an oracle system might need some kind of feed granularity. Like, "This oracle is doing a fine job on feed A, but feed B needs some attention."
Or maybe it's just all-or-nothing for a particular SMT and not feed granularity. Like, "This oracle is doing a fine job on SMT A, but not all of the data on SMT B is up to par, so they're disqualified from SMT B entirely."
Let's also look at how User-Feedback flows into the process. If you are a user trying to earn SMT, will you react in either scenario where a) Oracles are not including you (incorrectly) in their feed and b) where other are being included (incorrectly) in their feed. The answer is the user will react, likely publicly and on Steem, to both because they have an incentive to do so.
This feedback loop should strengthen Action/Event Feeds better than Price Feeds because of the difference in potential loss for individuals rather than social loss.
Sounds like the solution to what I'm describing is to limit/focus the number of related feeds per SMT. If there's a problem with a certain feed, it needs to be shed from the SMT or put on its own SMT.
Splinter early, splinter often.
Like, say you have a single SMT dealing with identity. It seems like two feeds might be good. One certifying validity, another certifying individuality. But in reality, it might be better to have two SMTs with their own feeds.
Since SteemFest conversations, it’s possible we should construct ability for second layer Oracles to act as QA on first layer Oracles. For instance if using Identity as an SMT rewards pool construction, an Oracle(s) should be paid/incentivized to identify accounts in the first layer that are not true single identities.
i was wondering about that.. so the same sort of Oracle used in Bitshares to determine the value of bit.USD is going to also determine whether someone has a real account or a fake one, ... are we talking about a crawler that just scans all parts of the comments section of a wordpress blog for example? and the n allows someone to use tokens on a wordpress blog??? and the oracle is for determining which accounts were primary accounts and not simply setup with similar names to fol the smart media token system? Like Youtube users who change their name to look like "Taylor Swift Vevo": to fool users into clicking links etc etc, this is to avoid stuff like that? To avoid mistaking user accounts on other websites??
Why not just ask the oracle to tel us the future? Do it alexnader the great style, march right into that temple, and grab the oracle by the throat and FORCE her to give us a winning prophecy!
"Steemit shall be undefeated in battle!"
yes, but the most important question remains unanswered...does the oracle bake cookies? :)
No the oracle is baked ON cookies.
Cool post and definitely a problem. To what extent decentralization can exist depends on the context and probably economic realities. Satellites are pricey after all. Upvoted and followed
Once it luanch not only oracle to adopt but almost all existing website
a well written article. Yes nothing stays the same as we are in a flux of change.
Resteemed