The Misestimation of Uncertainty in Affective Disorders

Maintaining estimates of why unexpected events happen, of why we are uncertain, allows humans to adjust their inference and learn as efficiently as possible.Humans adjust their learning in response to the levels of different types of uncertainty, including expected uncertainty, caused by probabilistic associations between observed events and what we are learning, and unexpected uncertainty, caused by changes in the thing we are learning.People who experience anxiety are less sensitive to changes in levels of unexpected uncertainty.Misestimation of the levels of uncertainty may lead to a tendency to be more influenced by negative events, which is believed to cause depression and anxiety.Understanding how the estimation of uncertainty is related to symptoms of anxiety and depression promises to enhance our understanding of these illnesses and may suggest novel therapeutic interventions. Our knowledge about the state of the world is often incomplete, making it difficult to select the best course of action. One strategy that can be used to improve our ability to make decisions is to identify the causes of our ignorance (i.e., why an unexpected event might have occurred) and use estimates of the uncertainty induced by these causes to guide our learning. Here, we explain the logic behind this process and describe the evidence that human learners use estimates of uncertainty to sculpt their learning. Finally, we describe recent work suggesting that misestimation of uncertainty is involved in the development of anxiety and depression and describe how these ideas may be advanced. Our knowledge about the state of the world is often incomplete, making it difficult to select the best course of action. One strategy that can be used to improve our ability to make decisions is to identify the causes of our ignorance (i.e., why an unexpected event might have occurred) and use estimates of the uncertainty induced by these causes to guide our learning. Here, we explain the logic behind this process and describe the evidence that human learners use estimates of uncertainty to sculpt their learning. Finally, we describe recent work suggesting that misestimation of uncertainty is involved in the development of anxiety and depression and describe how these ideas may be advanced. When we observe our environment, we observe it incompletely and intermittently. Our senses report aspects of the particular location we find ourselves in, but many of the most important processes we want to know about cannot be directly perceived. For example, if I stroke my cat, will it scratch me or not (Figure 1A , Key Figure)? If we knew whether the cat was feeling angry or friendly, then we would know whether it was likely to scratch us, but the internal state of the cat is hidden and so, if we want to know it, it must be inferred from our previous experience of how it has behaved. To complicate matters further, the hidden processes (see Glossary) we estimate will often change over time, so even if the cat was friendly the last time we stroked it, it may now be angry. Therefore, the challenge is how to make reasonable inferences about hidden, changeable processes when faced with such fundamental uncertainty about these processes. Failure to meet this challenge reduces our ability to accurately gauge the opportunities and costs afforded by the environment and, ultimately, to select the best course of action. One approach to mitigating the impact of uncertainty on our decisions is to identify its potential causes and use estimates of the extent of these causes to sculpt how we learn [1.Yu A.J. Dayan P. Uncertainty, neuromodulation, and attention.Neuron. 2005; 46: 681-692Abstract Full Text Full Text PDF PubMed Scopus (1078) Google we to learn why we not know what we are this we on how estimates of types of uncertainty, expected uncertainty and unexpected uncertainty (see of these and of may be used to improve learning, the evidence that human learners such and that with uncertainty estimation may be symptoms of such as anxiety and of this we on expected and unexpected Here, we describe these and of uncertainty of uncertainty in PubMed Scopus Google in and the PubMed Scopus Google of PubMed Scopus Google time we meet a cat, we not know whether it is likely to scratch of knowledge is and, in is to expected important is that estimation uncertainty of of we can it by more evidence the cat and what with expected uncertainty, the of knowledge is a fundamental of the that is not by more uncertainty when events (Figure in the and is to the is used when a is and expected uncertainty when it is more experience not expected uncertainty, which has in it uncertainty when the thing we are learning about changes (Figure in the uncertainty may be in the in in the by the cat is so we can be whether the of the cat has these unexpected uncertainty is often as an estimate of the that an will change between if we are how our cat and, a it we can that a change has a the are the of an is the that a change has over a time may change over allows the of uncertainty about the of the of may be their more and their impact on estimates of the environment more is likely to be a which are not in human learning. a more when how humans what are the hidden processes that are this we on expected and unexpected Here, we describe these and of uncertainty of uncertainty in PubMed Scopus Google in and the PubMed Scopus Google of PubMed Scopus Google time we meet a cat, we not know whether it is likely to scratch of knowledge is and, in is to expected important is that estimation uncertainty of of we can it by more evidence the cat and what with expected uncertainty, the of knowledge is a fundamental of the that is not by more uncertainty when events (Figure in the and is to the is used when a is and expected uncertainty when it is more experience not expected uncertainty, which has in it uncertainty when the thing we are learning about changes (Figure in the uncertainty may be in the in in the by the cat is so we can be whether the of the cat has these unexpected uncertainty is often as an estimate of the that an will change between if we are how our cat and, a it we can that a change has a the are the of an is the that a change has over a time may change over allows the of uncertainty about the of the of may be their more and their impact on estimates of the environment more is likely to be a which are not in human learning. a more when how humans what are the hidden processes that are that the cat was in a but it me when I stroked events when our understanding of the environment is less and the potential causes of a event may be to the of the cat that when stroked it, it may be that the cat is in feeling friendly, but to scratch on this or that the cat is feeling friendly and is now angry. the of these the event is caused by a probabilistic between the cat and its the cat may scratch even if it is feeling friendly not scratch if it is feeling angry. even whether the cat has or we cannot be of its of uncertainty is or uncertainty [1.Yu A.J. Dayan P. Uncertainty, neuromodulation, and attention.Neuron. 2005; 46: 681-692Abstract Full Text Full Text PDF PubMed Scopus (1078) Google and as the between an and a hidden process less For example, expected uncertainty would be if the cat on of the that it is in a and if it on of When how to to this of events, the the expected uncertainty, the less a about the hidden process we are in, and the more we to to estimate the state of the hidden process (see a of the of different of uncertainty on the of expected uncertainty about the of the cat that we its previous time to be of whether it is in a or a expected uncertainty would to make a on of on the of in the different of uncertainty how recent to more previous events are when learning about a hidden uncertainty, from a probabilistic between the hidden process and reduces the of events learners to on more previous events learning (i.e., use a learning in unexpected uncertainty the of by making it more likely that the process has observed (i.e., the use of a learning Here, the can be as the of an event to a is to it on the of the event as as the about the state and of the One approach that has used to the of events is to a of a and use the of the of the from to a particular event as a of the of the event of and in and PubMed Scopus Google the (see the of in an PubMed Scopus Google a of an its about the state of aspects of the as of a with to this that the of the about the of the that the process it is learning about may change over is a in this of the change in the from to the and can be as the to which the event changes the about the process that it to of this approach is in from this of the of unexpected and expected uncertainty and in on the in of is when is and when is the learning in response to uncertainty in the the of a learning is used when events have in the different of uncertainty how recent to more previous events are when learning about a hidden uncertainty, from a probabilistic between the hidden process and reduces the of events learners to on more previous events learning (i.e., use a learning in unexpected uncertainty the of by making it more likely that the process has observed (i.e., the use of a learning Here, the can be as the of an event to a is to it on the of the event as as the about the state and of the One approach that has used to the of events is to a of a and use the of the of the from to a particular event as a of the of the event of and in and PubMed Scopus Google Here, the (see the of in an PubMed Scopus Google a of an its about the state of aspects of the as of a with to this that the of the about the of the that the process it is learning about may change over is a in this of the change in the from to the and can be as the to which the event changes the about the process that it to of this approach is in from this of the of unexpected and expected uncertainty and in on the in of is when is and when is the learning in response to uncertainty in the the of a learning is used when events have why the cat may scratch is its has and, even it was feeling friendly, it is now angry. of uncertainty, is when the hidden process we are changes [1.Yu A.J. Dayan P. Uncertainty, neuromodulation, and attention.Neuron. 2005; 46: 681-692Abstract Full Text Full Text PDF PubMed Scopus (1078) Google is if the of the cat changes to if it is uncertainty the response to expected when learning about a we cannot so on previous it is more likely that the hidden process will have we have to our on more recent if the of the cat has unexpected uncertainty, we our about its on how it the last we stroked it, if it has unexpected uncertainty, we can a of events in our learning a how their in response to experience and, a about and to One of the is the learning of in the of and and in which a time is by a the between an event which has and what the expected to about the by a learning learning in is between and of the learning more if we to the on the can be the learning the as a of from the observed and in the of the previous with the learning as the When the learning is the more on the recent when it is the more on if an that events are by in the between and the hidden process then it be more influenced by events, and the use a learning (Figure if an the events are caused by changes in the hidden process then it be more influenced by recent events and use a learning (Figure Here, it is important to that the learning not estimate it that learning not is as a to how learning to changes in expected and unexpected of more which are to estimate uncertainty, are in more use their estimates of uncertainty to their learning in the of this we have the logic behind estimates of uncertainty to guide learning. have in which the learning was used to estimates of learning when uncertainty was the of in an PubMed Scopus Google learning of PubMed Scopus Google have learning the of PubMed Scopus Google as a response to the of and PubMed Scopus Google these the was used as a of learning it not describe how learners might whether learning be is to in which uncertainty is or more which are to their learning in response to uncertainty, have which we of the learning was by and in the of but not of PubMed Scopus Google and by of human and in PubMed Scopus Google of are to to changes in unexpected uncertainty by the with a (i.e., learning the of the and this estimate to adjust the learning of the this are in which of uncertainty are learning of learning by PubMed Scopus Google the of in a PubMed Scopus Google Finally, which of the environment, including estimates of of uncertainty, have most of these is the approach to and Scopus Google which can efficiently estimate and to changes in unexpected recent have which estimates of expected uncertainty the of in an PubMed Scopus Google in and the PubMed Scopus Google (see in the different in different with the of the from the to the To it is not which a more of how humans the of estimation of this we have the logic behind estimates of uncertainty to guide learning. have in which the learning was used to estimates of learning when uncertainty was the of in an PubMed Scopus Google learning of PubMed Scopus Google have learning the of PubMed Scopus Google as a response to the of and PubMed Scopus Google these the was used as a of learning it not describe how learners might whether learning be is to in which uncertainty is or more which are to their learning in response to uncertainty, have which we of the learning was by and in the of but not of PubMed Scopus Google and by of human and in PubMed Scopus Google of are to to changes in unexpected uncertainty by the with a (i.e., learning the of the and this estimate to adjust the learning of the this are in which of uncertainty are learning of learning by PubMed Scopus Google the of in a PubMed Scopus Google Finally, which of the environment, including estimates of of uncertainty, have most of these is the approach to and Scopus Google which can efficiently estimate and to changes in unexpected recent have which estimates of expected uncertainty the of in an PubMed Scopus Google in and the PubMed Scopus Google (see in the different in different with the of the from the to the To it is not which a more of how humans the of estimation of events can from a of causes of events may with an learning unexpected uncertainty and a learning expected to this is to whether the in a event is the (i.e., unexpected or not (i.e., expected of and in and PubMed Scopus Google but not and the to PubMed Scopus Google that of mitigating the impact of uncertainty on is to identify the potential causes of the uncertainty, estimate the extent of and then to use these estimates to sculpt how we the we the evidence suggesting that human learners and use such in the previous learning is more if a learning is used processes (i.e., when unexpected uncertainty is to when it is a of learning the of in an PubMed Scopus Google learning of PubMed Scopus Google of learning by PubMed Scopus Google have learning the of PubMed Scopus Google as a response to the of and PubMed Scopus Google of in PubMed Scopus Google of change the human Google have whether humans their learning when the of the process is as by the of learning, use a learning in to (i.e., when unexpected uncertainty is with is in the of in an PubMed Scopus Google learning of PubMed Scopus Google of learning by PubMed Scopus Google of in PubMed Scopus Google of change the human Google and have learning the of PubMed Scopus Google as a response to the of and PubMed Scopus Google of of this process have in the that with the of in an PubMed Scopus Google learning of PubMed Scopus Google of change the human Google of in of PubMed Scopus Google on learning in a Full Text Full Text PDF PubMed Scopus Google with a in from in the Full Text Full Text PDF PubMed Scopus Google of from an of work [1.Yu A.J. Dayan P. Uncertainty, neuromodulation, and attention.Neuron. 2005; 46: 681-692Abstract Full Text Full Text PDF PubMed Scopus (1078) Google that of the an estimate of is with of and 2005; PubMed Scopus Google of an inference PubMed Scopus Google on the of which that it the of and, their impact on (i.e., to an learning of the is with in between and in the and Full Text Full Text PDF PubMed Scopus Google suggesting that it would be to estimate of this with the of in previous have when was to when it was of learning by PubMed Scopus Google have learning the of PubMed Scopus Google as a response to the of and PubMed Scopus Google of in PubMed Scopus Google of in of PubMed Scopus Google to unexpected uncertainty, expected uncertainty learning that have expected uncertainty have used in which have to learn the the of an such are a with the of this process the expected of this of is that expected uncertainty of the and unexpected uncertainty in the of the can be in learning as expected uncertainty has of this learning of learning by PubMed Scopus Google of in PubMed Scopus Google to learning in PubMed Scopus Google in the human and and learning Full Text Full Text PDF PubMed Scopus Google in learning from an of uncertainty in PubMed Scopus Google work that the in the of a as that the learning in the human and and learning Full Text Full Text PDF PubMed Scopus Google that learning to the the this as evidence of the by which estimates of expected uncertainty is evidence that human learners to changes in unexpected uncertainty in the by of learning and evidence of a process expected suggest that humans estimates of the uncertainty of the associations are learning and use these estimates to their learning. work humans are to to changes in the unexpected and expected uncertainty of in When the of uncertainty estimation in as we it is to what the impact on might be if an was less to estimate the or of uncertainty learning. of the of uncertainty can be from the of learning expected uncertainty unexpected uncertainty would lead an to be less sensitive to recent events and to to important in less (Figure expected uncertainty unexpected uncertainty would lead to an of recent events and in (Figure Therefore, a of uncertainty misestimation is a ability to accurately the state of and, to select the most likely to or negative a the misestimation of uncertainty may a between the internal of depression and anxiety and the the such as to of uncertainty misestimation would be expected if the uncertainty of associations For example, as a tendency to the unexpected uncertainty of negative to would lead an to be more influenced by recent negative events, a that is of depression and anxiety to 2005; PubMed Scopus Google and in Full Text Full Text PDF PubMed Scopus Google as the of uncertainty, its may be may example, by the that expected uncertainty the change to the environment that unexpected uncertainty of I in by expected and unexpected uncertainty on learning that this of would lead to estimation of the state of the environment and the of uncertainty misestimation is that the the of uncertainty has of uncertainty the of in an PubMed Scopus Google have learning the of PubMed Scopus Google has about which of uncertainty has of and in and PubMed Scopus Google to learning in PubMed Scopus Google in the human and and learning Full Text Full Text PDF PubMed Scopus Google or has the of different of uncertainty different by making changes caused by unexpected uncertainty changes causes by expected uncertainty of learning by PubMed Scopus Google of in PubMed Scopus Google in learning from an of uncertainty in PubMed Scopus Google a the to which humans can accurately uncertainty to a when it is not has to be (see experience is a depression and anxiety of in the of and anxiety and Full Text Full Text PDF PubMed Scopus Google suggesting that learning processes are involved in the of these recent and in Full Text Full Text PDF PubMed Scopus Google P. of in the anxiety PubMed Scopus Google and in the of PubMed Google PubMed Scopus Google of and learning in anxiety and PubMed Scopus Google have in the evidence that learning processes are with these Here, we the aspects of this the of the misestimation of uncertainty in anxiety and learning in anxiety is by a and learning in to with evidence that are to associations and of such associations with P. of in the anxiety PubMed Scopus Google of and learning in anxiety and PubMed Scopus Google in the anxiety a 2005; PubMed Scopus Google has that and in response to PubMed Scopus Google with to anxiety in Full Text Full Text PDF PubMed Scopus Google to and a PubMed Scopus Google learning in depression is less and, a of in the evidence that are less sensitive to are and in Full Text Full Text PDF PubMed Scopus Google and in the of PubMed Google a a of to anxiety have in of a of the and about the Google with more work in the and of and a PubMed Scopus Google of the on anxiety and depression has that have of learning and that such as how learn associations or their to a of To work has related to the uncertainty of that of learning, such as whether are to their learning to changes in unexpected expected in learning to the levels of uncertainty is important when learning about hidden, processes. that aspects of the learning with depression and anxiety are related to the misestimation of the we work that evidence this and how such misestimation One approach to whether symptoms of anxiety and depression may be with misestimation of the of uncertainty is to how with their learning in response to changes in the levels of have learning the of PubMed Scopus Google has the to which anxiety is with learning to the of an this a of to have a of anxiety a learning that of which to learn the between and of the the associations in the the learning used by in the and a learning was used of the was that this in learning as anxiety with anxiety their learning less in response to changes in unexpected with the of in unexpected uncertainty [1.Yu A.J. Dayan P. Uncertainty, neuromodulation, and attention.Neuron. 2005; 46: 681-692Abstract Full Text Full Text PDF PubMed Scopus (1078) Google of and 2005; PubMed Scopus Google of an inference PubMed Scopus Google was to be when in the in the with the of this in this evidence that symptoms of anxiety are with a to changes in unexpected uncertainty, the of the and the between uncertainty estimation and symptoms (see by which the misestimation of uncertainty may impact symptoms of anxiety and depression is by the of in a that negative over of negative events, as is believed to be to anxiety and depression to 2005; PubMed Scopus Google and in Full Text Full Text PDF PubMed Scopus Google of to a of the and to PubMed Scopus Google in the of this the response to changes in uncertainty is to adjust the to which are influenced by to previous events (i.e., to use a or learning to the of the events with anxiety depression are more influenced by negative by events in the human and and learning Full Text Full Text PDF PubMed Scopus Google with to anxiety in Full Text Full Text PDF PubMed Scopus Google to and a PubMed Scopus Google suggesting that the uncertainty with the of events such that negative events are to be more would to estimates of the with and negative as a response to the of and PubMed Scopus Google whether such estimates by a in which to and negative and of the and, the unexpected of the such in more in the more to make the more more to which was in response to the was by learning and was such that their learning the and negative with a used was more and this learning evidence that human learners estimates of the unexpected of and negative and use these estimates to the to which their in response to a particular of of estimates of the with and negative events that the estimates may the negative anxiety and a have to the potential of misestimation of the of uncertainty in anxiety and to are and have on unexpected on expected uncertainty, suggest that of uncertainty estimation would to identify novel of the the of this has on the misestimation of uncertainty can changes in how an with their environment (see on the impact of the misestimation of that in uncertainty may such as response to of uncertainty in PubMed Scopus Google or with the of the PubMed Scopus Google it will be to the of these by uncertainty estimation processes in of with a of and are in the are which is that the of the world we to learn and our to directly observe many important aspects of it, make learning is evidence that strategy used to improve inference the causes of events, the of these causes and these estimates to sculpt how we learn from the and depression are with in learning, about events, an impact of negative over is of the and the of uncertainty may of these in the of and that understanding how this misestimation is now our and that the impact of that it be a accurately can human learners between expected and unexpected has used in which expected and unexpected uncertainty are the of how humans adjust their learning when this is not the is the misestimation of uncertainty to process by which levels of uncertainty are may be influenced by suggesting a potential by which impact would be to example, whether experience of was with of the unexpected uncertainty of negative is uncertainty in the and human work has a of in which with estimates of uncertainty, it is not how the estimate of uncertainty related to and negative are to changes in to anxiety that symptoms of anxiety with to changes in not evidence that this is to of unexpected uncertainty has in with and, may be of a of anxiety is the between misestimation of uncertainty and between a such as the of uncertainty, and symptoms is if it about the or a potential the this in which the process is by or interventions. accurately can human learners between expected and unexpected has used in which expected and unexpected uncertainty are the of how humans adjust their learning when this is not the is the misestimation of uncertainty to process by which levels of uncertainty are may be influenced by suggesting a potential by which impact would be to example, whether experience of was with of the unexpected uncertainty of negative is uncertainty in the and human work has a of in which with estimates of uncertainty, it is not how the estimate of uncertainty related to and negative are to changes in to anxiety that symptoms of anxiety with to changes in not evidence that this is to of unexpected uncertainty has in with and, may be of a of anxiety is the between misestimation of uncertainty and between a such as the of uncertainty, and symptoms is if it about the or a potential the this in which the process is by or interventions. would to on the in this was by a to and by the has from and and as uncertainty that of a between events and hidden For example, the cat of the time it is even when it is in a (see in the process that we cannot directly observe but that is a hidden process a the or how a will a of the between what and what was the learning is often as the of the the it is as the negative of the of the the of the and uncertainty that of a change in the thing that is For example, the of the cat changes (see in the the of a of unexpected related is a which is when a process changes (see in the