Angela Pretorius's Shortform

I'm curious about the long-term effects of eviction laws on homelessness rates.

I have a an intuitive argument (below) that removing eviction laws and making no-fault evictions really
easy would allow the housing market to clear and cause lower rents and (in the long term) a drop in
homelessness rates.

Does anyone here know whether there have been any empirical studies into the long-term effects of eviction
laws on homelessness and whether or not they have borne out the conclusion that eviction laws increase the
rate of homelessness? If not, are there any good theoretical arguments as to why eviction laws may or may
not increase homelessness rates?

Here is my argument:

Most rental accommodation stands empty for long periods of time. Why would landlords prefer to let rooms
stand empty than to lower their rents and get their rooms filled? And why do so many landlords refuse to
rent to people of certain demographic groups? For example, I many know of landlords who refuse to rent to
construction workers, black people or people who claim benefits and would rather lose money by letting
their accommodation stand empty. Even where there are anti-discrimination laws to prevent this from
happening, in practice the anti-discrimination laws are impossible to enforce.

The reason is that every time a new tenant moves in, the landlord has to trust that the new tenant will pay
their rent and won't cause property damage or complaints from the neighbors. Because it is so difficult to
evict a 'problem' tenant, the landlord will not rent to demographic groups that he/she perceives as
containing a higher proportion of problem tenants.

In addition, because tenants of lower socioeconomic status are more likely to become 'problem' tenants,
landlords deliberately keep their rents high to discourage those of lower socioeconomic status from renting
from them. As an example, £1000 per month for a property and having it stand empty half the time brings in
an average of £500 per month, whereas charging £800 per month for a property and having it occupied 90% of
the time brings in £720 per month, but if the sort of tenants who can afford £800 per month are more likely
to become problem tenants than the sort of tenants who can afford £1000 per month, and if eviction laws
make the cost of taking in problem tenants very high, then it will be more profitable to charge £1000 per
month.

One other point that needs to be made is that the people who are most at risk of homelessness tend to be
those who are impulsive/heavily discount the future. Imagine a tenant who values the pleasure of engaging
in a certain problem behavior at 10 units and values being evicted and made homeless now at -50 units, but
discounts the future heavily and only values being made homeless in five months time at -5 units. If there
are no eviction laws and engaging in this problem behavior will lead to immediate eviction and
homelessness, then this tenant will decide not to engage in this problem behavior and will not become
homeless. On the other hand, if there are strict eviction laws and engaging in this problem behavior will
lead to a 50% chance of eviction after five months of court cases then this tenant will decide to engage in
the problem behavior.

I think that psychosis can be characterised as a failure of inner alignment.

Let me explain with two examples.

1. We all seek status, but we obviously don’t have a genetically hardwired status classifier in the brainstem. Instead the rest of the brain figures out what our social status is. When something boosts our estimated status the ventral tegmental area in the brainstem sends out a dopamine reward signal. If the brain’s inner alignment mechanisms fail then the rest of the brain maximises its dopamine reward by convincing itself that it is the messiah.
2. The parts of the brain which detect threats get rewarded whenever they detect a threat so that they have an incentive to be vigilant to threats. If inner alignment fails then the threat detector circuits try to maximise their dopamine reward by hallucinating ghosts or snakes or whatever.

Many antipsychotics are dopamine antagonists. Dopamine antagonists don’t fix the underlying alignment failure but they trade off less psychosis for more apathy.

##AGI might not be conscious

###I don’t think that feedforward networks can be conscious

Integrated information theory does not allow for conscious feedforward networks.

###There is no good reason to expect AGI to be a recurrent neural network

Transformer models like GPT are basically feedforward networks [to do: check that this is correct]. Recurrent neural networks have had some successes but they are difficult to train. Whereas feedforward networks can be trained quite effectively using back-propagation. Biological neural networks are unable to implement back-propagation and instead rely on a variant of Hebbian Learning. To implement a recurrent neural network trained via Hebbian Learning on a computer wouldn’t necessarily be the best use of computational resources. AGI may very well be a feedforward network trained using back-prop.

###Anthropics All of human consciousness is about to be massively dwarfed by an enormous explosion of superintelligent, highly conscious AIs. And by some incredible co-incidence I happen to be human???

It just seems so implausible. An explosion of unconscious or minimally conscious superintelligent AIs is much more plausible.